A method and apparatus for loop-filtering the image data

 

(57) Abstract:

The invention relates to television, in particular to the processing of the image data, and in particular to a method and apparatus for loop-filtering the image data. The technical result - the effect of quantization that occurs when encoding and decoding the image data. The invention consists in that way the loop filtering includes the steps of selection flag that indicates whether the data require image loop-filtering using the distribution coefficients of the inverse conversion (CPC) is inversely converted image data and the displacement vector indicating the difference between the previous frame and the current frame. Corresponding to the flag data of the image is then pre-filtered in a certain way, if the flag specifies whether the loop-filter. The use of flags and the adaptive filter reduces the effect of quantization and allows you to reduce the amount of calculations required for filtering. In addition, filtering may be performed by parallel processing without dividing, and multiplying that allows the use of simpler hardware. 2 C. and 29 C.p. f-crystals, 8 ill.device for loop-filtering the image data, to reduce the effect of quantization.

As a rule, standards for encoding images, such as MPEG International Organization for Standardization (ISO) and in the format H.263 recommended by International Telecommunication Union (ITU), include block motion estimation and discrete cosine transform blocks (DCT). Most coding standards image uses DCT to blocks of 8x8 pixels for packaging information with a small number of transform coefficients. This block diagram of the DCT based on the properties of the local spatial correlation of images.

However, when restoring the image data on a block basis, there is significant distortion of the image, for example, there are blocking artifacts at the borders of the block, corner emissions at the point of intersection of the blocks and low-frequency noise transients from the edges of the image ("ringing"). This is because the image data pass through the DCT blocks of 8x8 pixels to quantization. This distortion of the data is significant with a high degree of image compression.

Blocking artifacts are caused by the noise of the mesh along the boundary of block in a relatively homogeneous region. The noise becoming visible on the screen. Thus, the edges between the blocks become visible. The angular emission is formed where the corners of the blocks of 8x8 pixels. In addition, low-frequency noise is a typical phenomenon Gibbs, due to rounding, when the coefficients of high frequency components of the DCT quanthouse for maximum compression of the image. Because of the low frequency noise may be overlapping images with pre-defined intervals.

It was suggested several ways to reduce blocking artifacts, the angular emission and low-frequency noise generated by block coding. The way a single adaptive filter (AAF) to reduce the effect of quantization of the decompressed JPEG images was proposed by D. L. Lee, S. Kim, and C. Park, in "Reduction of blocking effect in JPEG images for adaptive signal processing" [Press IEEE Trans. "Image processing", 1997]. There is a method that uses a two-dimensional filter to reduce the blocking artifact, whereas one-dimensional (1-D) filter is used to reduce the speed of the interference (see "Nonlinear spatial version of the final image processing block coding", I is receiving design theory for convex objects, described in the "Projection spatially adaptive reconstruction of block-converted compressed images", IEEE Trans. on Image processing, Vol. 4., N 7, pp. 896-908, July 1995. In addition, published the article "Algorithm release for compressed images in JPEG format, using the representation of elementary waves", IEEE Trans. System Video Technology, Vol. 7, N 2, pp. 433-437, 1997.

However, the above methods require complex calculations.

In order to solve the above technical problem, the present invention is a method and device loop-filtering image data capable of encoding with a low data transmission rate in bits per second without complex calculations and mitigating the effects of quantization, such as blocking artifacts, the angular emission and low-frequency noise caused by the block decoding.

According to one aspect of the invention a method is proposed loop-filtering to reduce the effects of quantization that occurs when encoding and decoding image data, comprising the following stages: allocation flag indicating the need to use the loop-filtering the image data using raspredeleniya, pointing to the difference between the previous frame and the current frame and the filtered image data corresponding to the flag, pre-defined way, if the flag specifies whether the loop-filter.

At stage (a) it is preferable that the flag stood out from the field of discrete cosine transform (DCT) of each block, if the frame image, which should be subjected to the loop filter is an internal frame, and the residual signal and the reference frame, if the frame image, which should be subjected to the loop filter is an intermediate frame.

Preferably, the flag contained the lockout flag indicating the need to reduce blocking artifacts at block boundaries, and the interference flag indicating the need to reduce low-frequency noise on the edges of the image.

In the case where the image data represent the inner frame, the flag lock the inner frame preferably includes a lock flag horizontally (FBG) and the lockout flag vertically (FBW), with a selection of FBG and FBW inner frame includes the following stages: calculation of the coefficients of the DCT block of 8x8 pixels Parham left corner of the block 8x8 consisting of 64 pixels, the pixel B is located to the right of the pixel A and pixel C is located below the pixel A, FBG and FBW are set to "1", pointing to the need for the loop-filter, if the coefficients of the pixels is not equal to zero; VBW is set to be "1", pointing to the need for the loop-filter, if only the upper row of the inverse quantized block 8x8 is not equal to zero, and VBG is set to "1", pointing to the need for the loop-filter, if the leftmost column of the inverse quantized block 8x8 is not equal to zero.

When the selection flag of interference from the inner frame, assuming that the pixel A (coefficient DK) is located in the upper left corner of the 8x8 block consisting of 64 pixels, the pixel B is located to the right of the pixel A and pixel C is located below the pixel A, the interference flag is preferably set to "1", pointing to the need for loop-filtering when any pixel, in addition to pixels A, B and C are inverse quantized block 8x8 has a coefficient not equal to zero. Flag lock the current intermediate frame contains preferably the lockout flag horizontally and lockout flag vertically, and, assuming that the reference frame includes a predetermined number of current intermediate frame, is the unit displacement X, the allocation FBG and FBW block Acthe current intermediate frame includes the following stages: analysis of the degree of overlap between the block displacement X and the reference blocks; bitwise logical operation "And" FBG and FBW supporting blocks having a predetermined number of overlapping pixels, and the establishment of the bitwise "And" as FBG and FBW block Acin the plane of the current video object.

Preferably the selection flags of the block horizontally and vertically in the block Acthe current intermediate frame further includes the step of setting the kill bit reference macroblock as a flag block of the current macroblock, if the displacement vector of the current macroblock is a skip macroblock having zero displacement vector.

Preferably the selection flags of the block horizontally and vertically in the block Acthe current intermediate frame further includes the step of filling the data block insignificant information after copying the block flag.

Preferably, the selection flag of interference in the block Acthe current intermediate frame when the frame of the c in state "1" if any variable component of the coefficients of the inverse conversion (CPC) of the residual signal of the intermediate 8x8 blocks is not equal to zero, and setting the flag interference in state "0" if all variables are components of the inverse transform coefficients are zero; installation of interference flag to "1" when the current block is a prediction mode of the block 8x8 for transmitting the displacement vector in the macroblock.

Preferably in the case where video data represent an intermediate frame, the flag of the interference of the intermediate frame includes the first flag interference (FPO) and the second flag interference (FP), and the selection of the first and second flags interference FPO and FP in block Acthe current intermediate frame includes the following stages: setting the first flag interference FPO of the current block is in state "1", if the constant components of the coefficients of the inverse transform of the residual signal of the intermediate 8x8 macroblocks is not equal to zero; otherwise, setting the FPO of the current block is Acin state "0", and setting the second flag interference FP of the current block is in state "1" if any variable component of the coefficients of the inverse transform of the residual signal of the intermediate Cadro is about, provided that the specified block constituting a frame to be loop-filtered is the block I and block adjacent to the block I, block J, the difference between the current block I, and the quantized previous unit I does not exceed the predetermined value and the difference between the current block J, and the quantized previous unit J does not exceed the predetermined value, filtering the release stage filtering is not performed.

Preferably the filter release to reduce blocking artifacts comprises the following steps: performing a first filtering for changing a predetermined number of pixels in the horizontal boundary block between blocks I and J, if the flags interference blocks I and J indicate the absence of the need to reduce low-frequency noise and both FBG blocks I and J indicate the need for reduction of blocking artifacts; comparing the difference between two adjacent pixels along the boundary of block quantization index N.263 QP, if both flags interference blocks I and J indicate the absence of the need to reduce low-frequency noise and at least one of FBG blocks I and J indicates no need for reducing the blocking artifacts, or the mA; change pre-defined number of pixels that is less than this number for the first filter if the difference between the two pixels does not exceed QP.

Preferably the filter release to reduce blocking artifacts comprises the following steps: performing a first filtering for changing a predetermined number of pixels in the block between blocks I and J, if the interference flag of the block I indicates no need for reducing low-frequency noise and both units I and J are internal blocks, comparing the difference between two adjacent pixels along the boundary of block quantization index N.263 QP, if the interference flag of the block I indicates no need for reducing low-frequency noise and at least one of FBG blocks I and J indicates no need for reducing the blocking artifacts, and performing a second filtering for changing a predetermined number of pixel values that is less than this number for the first filter if the difference between two pixels is less than the QP; the first filter if the flag interference of the block I indicates no need for reducing low-frequency noise, and no block I, of (CPC) the residual signal of the block I and block J and both FBG blocks I and J are in the state "1", performing a second filtering, if the interference flag of the block I indicates no need for reducing low-frequency noise, and no block I or block J are inner blocks; there is a constant component of the coefficient of the inverse conversion (CPC) of the residual signal of the block I or J, and at least one of FBG blocks I and J is equal to zero, and performing a second filtering, if the interference flag of the block I indicates the need to reduce low-frequency noise.

Preferably the filter release to reduce blocking artifacts comprises the following steps: performing a first filtering for changing a predetermined number of pixel values in the block between blocks I and J, if the second flag interference FP of the block I indicates no need for reducing low-frequency noise, no block I or block J are inner blocks and both FBG blocks I and J indicate the need for reduction of blocking artifacts, a comparison of the difference between two adjacent pixels along the boundary of block quantization index N.263 QP, if FP of the block I indicates no need for reducing low-frequency noise; no block I or block J are inner blocks and, for me; performing a second filtering for changing a predetermined number of pixel values that is less than this number for the first filter if the difference between two pixels is less than the QP; the first filter, if FP of the block I indicates no need for reducing low-frequency noise; no block I or block J are inner blocks; at least one of the FPO blocks I and J is in state "1" and both FBG blocks I and J are in the state "1", execution of the second filter, if FP of the block I indicates no need for reducing low-frequency noise; no block I or block J are inner blocks; at least one of the FPO blocks I and J is in state "1", and at least one of FBG blocks I and J is not in the "1" state; and performing a second filtering if FP block I is not equal to zero.

Preferably, provided that six pixels on the boundary of the horizontal block between blocks I and J are denoted as A, B, C, D, E and F, where the pixels C and D is the nearest pixels on the boundary of the horizontal block, the pixels A and F represent the most distant pixels, and the pixels B and D - medium; the first filter is a low pass filtering to the pixel C and D are replaced by C=C+(D-C)/4 and D=D+(D-C)/4, if the absolute value of the difference between pixels C and D is less than QP N.263.

Preferably, the operation of reducing low-frequency noise is at a stage of filtering, if the interference flag indicates whether to perform the filtering action; this operation is not performed if the interference flag indicates no need for reducing low-frequency noise, the filtering operation of the low-frequency noise includes the following steps (a) detecting horizontal and vertical edges of the image data and (b) performing a two-dimensional (2-D) adaptive filtering of the signal in the 8x8 block, which must be reduced low-frequency noise. In step (a) to detect horizontal edges, provided that the pixel [m, n] is the current pixel, a pixel [m] [n+1] is located to the right of the pixel [m, n] and the pixel [m][n-1] is located to the left of the pixel [m, n], the difference between the pixel [m][n] and the pixel [m] [n+1] is expressed as A1, the difference between the pixel [m][n] and the pixel [m][n-1] is expressed as A2; QP - rate quantization H. 263, when (and(A1 > QP) and (A2>QP)) or (A1>2QP) or (A2>2QP); the current pixel is detected as an edge pixel, and the display region, the edge [m][n] becomes equal to 1. In addition, the step (a) to detect vertical edges, provided h is is pixel [m-1][n]; the difference between the pixel [m][n] and the pixel [m+l][n] is A'1; the difference between the pixel [m] [n] and the pixel [m-1][n] is A'2. Thus QP is the quantization index H. 263, satisfied when the conditional expression ((A'1>QP) and (A'2>QP)) or (A'1>2QP) or (A'2>2QP), the current pixel is detected as an edge pixel, and the display region, the edge [m][n] becomes equal to "1". At stage (b) adaptive filtering of the signal box filter with four inputs used for 8x8 block to determine whether the Central pixel of the filter window edge pixel, and the filtering is performed if the Central pixel is an edge pixel; otherwise, it is a weighted filtering.

Preferably the filter to suppress low-frequency noise on stage filtration stage is not performed if the difference between the current block I and the quantized previous unit I is not greater than a predetermined value, and the difference between the current block J, and the quantized previous block J is not greater than a predetermined value.

Preferably the filter to reduce low-frequency noise is performed on the filtration stage, if the interference flag of the block I indicates the need for such filtering; in atomnogo noise is, when the block I is an internal block, when the block I is not the indoor unit and is in the prediction mode of the 8x8 block and when the block I is not the indoor unit and is not in the prediction mode of the block is 8x8, and FP of the block I indicates no need for the filter to reduce low-frequency noise, and the filtering is not performed when FP of the block I indicates no need for the filter to reduce low frequency noise.

Preferably the filter to reduce low-frequency noise includes the following steps: (a) detecting horizontal and vertical edges of the image data and (b) performing a two-dimensional (2-D) adaptive filtering of the signal in the block 8x8, from which must be removed low-frequency interference. At the stage of (a) detecting horizontal edges, provided that the pixel [m,n] is the current pixel, a pixel [m][n+1] is located to the right of the pixel [m,n] and the pixel [m] [n-l] is located to the left of the pixel [m,n], and the difference between the pixel [m][n] and the pixel [m][n+1] is equal to A1, the difference between the pixel [m][n] and the pixel [m][n-1] is equal to A2, QP is a quantization index H. 263, when conditional expression is satisfied ((A1>QP) and (A2>QP)) or (A1>2QP) or (A2>2QP), the current pixel is detected by the vertical edges, assuming that the current pixel is pixel [m,n], pixel [m+1][n] is located above the pixel [m, n], pixel [m-1][nl located below the pixel [m, n], the difference between the pixel [m][n] and the pixel [m+1] [n] is A'1, the difference between the pixel [m][n] and the pixel [m-1] [n] is A'2, and QP is the quantization index H. 263, satisfied when the conditional expression ((A'1>QP) and (A'2 >QP)) or (A'1>2QP) or (A'2>2QP), the current pixel is detected as an edge pixel, and the display region, the edge [m] [n] becomes equal to "1". In addition, at the stage (b) adaptive filtering of the signal filter box with 4 inputs is applied to the 8x8 block to determine whether the Central pixel of the filter window edge pixel; filtering is performed if the Central pixel is an edge pixel; otherwise, it is a weighted filtering.

It is preferable to reduce the angular emission generated at the point where the corners of four blocks, when the image data are subjected to block coding and decoding, the way the loop filter further includes the following steps: (a) detection of the angular emission of a block of 8x8 inverse quantized image data and (b) the compensation of the detected angular emission. At the stage of (a) about, 2] and [3] represent the values of the pixels A, B, C and D, respectively, the average number of pixel values is set as (A+B+C+D+2)/4, the difference between each pixel value to the average value is compared with the quantization index N.263 QP, and the number of pixels with angular emission collected for detection of the angular emission if the difference is greater than the QP. In addition, the phase compensation (b), provided that A1and A2the pixels adjacent to the pixel A, and A3- the pixel diagonal to the pixel A in the same block of pixel A, A', A'1and A'2- compensated pixel values A, A1and A2accordingly, when the detected only one candidate angular emission and the detected candidate pixel A, the difference between the pixels A and A3less than 3QP/2, the compensation of the angular emission is performed using the following equations:

A' = (4A+B+C+2D+4)/8;

A'1= (A'+3A1+2)/4;

A'2= (A'+3A2+2)/4,

and, when the number of candidates is greater than two, selects the candidate who has the most contrast (A3+ B3+ C3+ D3+ 2)/4, and compensation of angular emission is performed at this point as it is done in the case of only one candidate.

BRIEF DESCRIPTION OF DRAWINGS

The above is Ianto embodiment of the invention with reference to the drawings, which presents the following:

Fig. 1 is a block diagram of a device for the loop-filtering the image data, that reduce the effect of quantization of a block of image data;

Fig. 2 - decoder structure corresponding to N. 263+ using the 8x8 block of coefficients of discrete cosine transform (DCT), inverse quantized and inverse quantizer decoder;

Fig. 3 is an illustration of the relationship between 8x8 block (Awith) in the inner frame and the adjacent blocks of the reference frame;

Fig. 4 is an example illustrating the allocation of block flag horizontally (FBG), block flag vertically (FBW) and interference flag (AF) in the plane of the intermediate video (CEA);

Fig. 5 is a representation of the boundaries of the block and the arrangement of the pixels adjacent to the boundary block, to illustrate filtering implemented by the filter release, to reduce blocking artifacts;

Fig. 6A is an illustration of image edges, which are the angular emission due to the quantization;

Fig. 6B - angular emission generated by the quantization;

Fig. 6C - coordinates for the points of the angle for angular emission;

Fig. 7A is a representation of a kernel two-dimensional (2-D) adaptive filter signal (ASF);

DESCRIPTION OF THE PREFERRED OPTIONS

In Fig. 1 shows a device 10 for loop-filtering the image data to reduce the effect of quantization that occurs when the image data block formed by the encoding and decoding implemented by the system with the General encoder and decoder. The device 10 includes a block 100 allocation flag, filter 110 release, the compensator 120 angular emission and the filter 130 to reduce interference. Block allocation flag 100 allocates a flag using the distribution coefficients of the inverse conversion (CPC) inverse quantized image data and the displacement vector indicating the difference between the previous frame and the current frame. The flag is information that indicates, does the decoded image to the loop-filter, and consists of block flag and the flag of interference. The lock flag indicates the need to reduce blocking artifacts at the boundaries of the block, and the interference flag indicates the need to reduce low-frequency noise (low frequency noise) from the edges of the image. The lockout flag consists of FBG indicating the execution of the loop-filtering pixels adjacent blocks along the horizontal edge of the block, and FBW indicating the execution of the CEC in the inner and intermediate frames. The present invention proceeds from the fact that each intermediate frame consists of a P frame and the frame of the RV. The distribution coefficients of the CPC inverse quantized image data used for the allocation flag in the internal frames. When the selection flag in the intermediate frames is the displacement vector indicating the difference between the previous frame and the current frame. A detailed description of this process is shown below.

The filter 110 release checks the flag block allocated a block of 100 allocation flag using one-dimensional (1-D) low pass filters (LPF) for filtering horizontally and vertically, carrying, thus filtering the release of the image data.

The compensator release 120 detects the angular emission data that passed through the filter 110 release, and compensates the detected angular emission.

The filter 130 suppress interference checks the flag interference allocated block allocation flag 100 using a two-dimensional adaptive filter signal (APS) in the noise filter and angular emission data. The main idea of the way the loop-filtering the image data in accordance with the present invention consists in advancemenu.

It also takes into account the subjective image quality, maximum signal-to-noise and computational complexity. In particular, if the main idea is implemented using software and hardware, the computational complexity is a crucial factor. Analyzed the distribution of the coefficients of the inverse transformation to highlight information about the blocking artifacts and the low-frequency noise in each 8x8 block. The low pass filter and two-dimensional adaptive filter signal (APS) adaptive image applied to each 8x8 block using flags blocking and interference.

In the present invention are of two varieties: the flag of interference that are installed in a single interference flag (AF) in the first version, and installed in the form of two, i.e. first and second, flags interference FPO and FP, in the second embodiment. If further description no link on the first or second option, it means that used both. In addition, in the case when using only the first or only the second option, the appropriate option necessarily mentioned.

The following describes the selection flag of blocking artifacts and interference flag using block Obi to reduce the effects of quantization in the format H.263+, defined two types of flags: the flags block and flags interference. Flags blocking and flags interference allocated from the zone COP of each 8x8 block in the inner macroblock (MB). In addition, flags intermediate macroblocks are calculated from the residual signal flags and the reference video object plane (TNA).

1.1. Information flags for internal macroblock

The distribution of the coefficients of the inverse transform (DCT coefficients after inverse quantization) is determined in the decoder. In Fig. 2 presents a block diagram of the decoder N. 263+ DCT coefficient after the inverse transform. In block inverse transform 8x8 in Fig. 2 the coefficients a, B and C are used for decision-making on the block flag and the flag of interference.

If only the coefficient in position A (constant component) among the 64 DCT coefficients of the 8x8 block shown in Fig. 2, null, 64 encoded pixel block 8x8 have the same values in the spatial domain. Thus, the block having a fixed component only, can create artifacts blocking horizontally and vertically. In this case, the flags block horizontally and vertically of the block are set at "1".

If the coefficients in the far left column are non-zero, eight pixels in each row have the same value in the spatial domain. This unit can create artifacts lock horizontally, so that the lockout flag is horizontally installed in the state "1".

In the first embodiment, the interference flag is set to state "1" if the coefficients are not zero, exist in positions other than A, B and C in Fig. 2. In the second case, the first flag of interference FPO is set to be "1" if the coefficients are not zero, exist in positions other than except A, B and C in Fig. 2.

The presence of these high-frequency coefficients indicates that the block includes the edges of the image. Therefore, the unit generates a low-frequency noise (low frequency noise) on the edges of the image due to rounding of the high-frequency coefficients. In the second embodiment, the second flag interference FP internal macroblock is not used.

1.2. Dissemination of the flag on the intermediate macroblock (MBP)

In the first VariCam vectors move. In addition, the residual signal of the intermediate block is used to define the flag of the intermediate block. Distribution block flag from the reference frame to the intermediate frame is as follows.

In Fig. 3 shows the relationship between 8x8 block (Acin the intermediate frame and the adjacent blocks of the reference frame. The passage of the lockout flag is described by the vectors of displacement MVxand MVyin the following way. In Fig. 3 ArBrCrand Drrepresent blocks of the reference frame. Ac- block in the current intermediate frame and the X - block move Acthat is measured using the displacement vector (MVx, MVy). It first examines the degree of overlap of the displacement vector X and the reference block. If you use only the units in which the area covered by the intermediate unit displacement X and the reference block is wider than the 2x2 pixels, you can calculate the blocking flags horizontally and vertically of the current block is Awithbitwise method using the Boolean operations "And" for FBG and FBW thrust blocks that overlap the block X is estimated from the displacement vector.

For example, if MVx= 5 and MVySUB>rand Dr. Here, all four overlapped region is wider than 2x2 pixels. Thus, FBG and PBW of the current block is Accan be calculated from the four supporting blocks ArBrCrand Dras shown in Fig. 4. Here mark & indicates a bitwise logical "And".

The following describes the interference flag (AF) in the intermediate frame. First of all OP reference block Acis set to be "1", if the ratio of the inverse transform of the residual signal in the 8x8 block of the intermediate frame is not equal to zero. Transfer the four vectors of displacement for one macroblock is determined as the prediction mode of the block is 8x8. The prediction mode of the block 8x8 relates to a busy area with a high-frequency component. Thus, it is determined whether the 8x8 block in the prediction mode, and AF unit with the prediction mode 8x8, is set to be "1".

As for the second option, then the locking flags block in the reference frame (the image I or image P or a portion of the image P improved PB frames) apply to the following intermediate macroblock using the vectors of displacement. In addition, the residual signal PR is a skipped macroblock (COD-1), the blocking flags six blocks (four for Y, one for Cband one for Crin the reference macroblock is duplicated in the corresponding blocks of the current macroblock.

Distribution block flag of the reference frame to the intermediate frame is the same as in the first embodiment. However, to calculate FBG and PBW of the current block is Acthe logical operation "And" applies to blocks whose overlap block X - wider than a 1x1 pixel, unlike the first variant, in which the logical operation "And" applies to blocks whose overlap block X - wider than 2x2 pixels. For example, if MVx-5 MVyis 3.5 estimated according to the translation vector of the block X is covered by four support blocks ArBrCrand Dr. There are four covered a wider area of 1x1 pixel.

The following is an explanation of interference flag in the intermediate frame. It first examines the coefficient of the inverse transform of the residual signal in the block 8x8 intermediate macroblock. FPO block Acis set to be "1" if the constant component of the coefficient of the inverse transform is not equal to zero and all other components equal to zero. In addition, FP pornog is of the residual signals is not equal to zero. The prediction mode 8x8 blocks is taken into account when filtering for noise reduction.

2. The way the loop-filtering using flag

The filter 110 release, the compensator 120 angular emission and the filter 130 suppression are described in detail below.

2.1. Filter release to reduce blocking artifacts

One-dimensional low pass filter to reduce blocking artifacts works in "strong" or "weak" modes on the horizontal and vertical boundaries of the unit depending on the position of the flags lock. As is customary in most release options for reducing blocking artifacts, we computed the information about the edges of the invention, and the lowpass filter is an adaptive on the basis of data on the edges of the image. However, the method of release in accordance with the present invention uses the above-described flag lock, so there is no need for edge detection image, which requires a large amount of calculations.

The current 8x8 block, which is subject to processing, and related blocks shown in Fig. 5. If both FBG blocks BLOKE and _J is set to "1", 7-element lowpass filter with coefficients

The filtering algorithm release horizontally in the first embodiment is expressed as follows:

if (BLOCK I == coded && _J HE coded) filtering the release fails:

if(internal frame){

if (FP BLOCKA == 0 && f FP _J == 0){/*NO interference*/

if (FBG BLOCKA == 1 && FBG _J == 1)

strong filtering release

otherwise,

weak filtering release);

}

if (frame P or PB) {

if (FP BLOCKA == 0) {/*NO interference*/

if (_J == internal||_J == internal){

if (FBG BLOKE ++ 1&&FBG _J == 1)

strong filtering release;

otherwise,

weak filtering release;

}

otherwise {

if (constant component of the residual GST BLOKE or _J){

if (FBG BLOCKA == 1 FBG _J == 1)

strong filtering release;

otherwise,

weak filtering release;

}

}

}

otherwise, /*low noise*/;

weak filtering release;

)

In the explanation of this algorithm, assuming that the unit includes a frame to be loop-filtering - this UNIT IS OCOM I does not exceed a specified amount, and the difference between the intermediate current _J and the previous quantized _J also does not exceed the preset value, filtering the release is not executed.

If the frame should be subjected to the loop-filter, the inner frame when the filter release to reduce blocking artifacts when both flags interference BLOKE and _J indicate that the reduction of low-frequency noise is not necessary, and both FBG BLOKE and _J indicate that the reduction of blocking artifacts is necessary, the first filter (strong filtering release) to change the number of values of pixels in the horizontal boundaries of the block between BLOXOM and _J.

If both flags interference BLOKE and _J indicate that there is no need to suppress low-frequency noise and at least one FBG BLOKE and BLAKE J indicates that the reduction of blocking artifacts is not necessary or when at least one of the flags interference BLOKE and _J indicates that the reduction of low-frequency noise is necessary, and the difference between two adjacent pixels in the horizontal boundaries of the block is less than the quantization index N. 263 (QP), is the second filtering is ation.

In addition, filtering the pixels at the border of the vertical block is the same as filtering pixels in the horizontal boundaries of the block, using FBW.

If the frame should be subjected to the loop-filter, the intermediate frame (frame P or PB), when filtering the release to reduce blocking artifacts when the flag interference BLOKE indicates that there is no need to suppress low-frequency noise, one of the blocks - BLOCK and _J is an internal block and both FBG BLOKE and _J indicate the need for reduction of blocking artifacts, is first filtered to change the number of values of pixels in the horizontal boundaries of the block between BLOXOM and _J.

If the flag is interference BLOKE I indicates that there is no need to suppress low-frequency noise, one of BLOKE and _J is internal, any of FBG BLOKE and _J indicates that there is no need to reduce the blocking artifacts, and the difference between two adjacent pixels

the horizontal boundaries of the block is less than the quantization index N.263 (QP), is the second filter to change the values of the pixels whose number is smaller than the first filter.

When the flag interference BLOKE indicates that there is no need to suppress low-frequency noise, no BLOCK or _J are not internal and at least one of FBG BLOKE and _J is in the state "0" is the second filtering.

When the flag interference BLOKE not equal to zero, is the second filtering.

Assuming that six pixels on the boundary of the horizontal block between BLOXOM and _J pixels A, B, C, D, E and F, in which the pixels C and D is the nearest pixels on the boundary of the horizontal block, the pixels A and F - pixels furthest from them, and the pixels B and D is the average of the pixels within the first filter is a 7-element filter with coefficients (1,1,1,2,1,1,1) for the six pixels.

The second filtering is performed on the pixels C and D. In this case, if the absolute value of the difference between pixels C and D is less than QP N.263, two pixels C and D are replaced as C = C+(D-C)/4 and D=D+(D-C)/4.

The filtering algorithm release horizontally in the second embodiment is expressed as follows:

if (BLOCK I == coded && _J NOT encoded) filter release otsutst BLOCKA == 1 && FBG _J == 1)

strong filtering release;

otherwise,

weak filtering release};

}

otherwise, a weak filtering release;

}

if (frame P or PB) {

if (FPO BLOCKA == 0) {/*NO interference*/

if (_J == internal||_J == internal){

if (FBG BLOKE ++ 1&&FBG _J == 1)

strong filtering release;

otherwise,

weak filtering release};

}

otherwise {

if (FPO BLOCKA==1||FPO _J==1){

if (FBG BLOCKA==1&&FBG BLOCKA==1)

strong filtering release;

otherwise,

weak filtering release};

}

}

}

otherwise, /*low-frequency interference*/;

weak filtering release;

)

To explain the algorithm, let us note that, if the block constituting a frame to be loop-filtered is BIOCOM, the block adjacent to BIOCOM is a _J, the difference between the current BIOCOM and the previous quantized BLOXOM I does not exceed a specified amount, and the difference between the current _J and the previous quantized _J does not exceed a specified amount, filtration zablokirovali release to reduce blocking artifacts, when the flags FPO BLOKE I and _J indicate no need for reducing low-frequency noise, and FBG BLOKE and _J indicate the need for reduction of blocking artifacts, the first filter (strong filtering release) to change the number of values of pixels in the horizontal boundaries of the block between BLOXOM and _J.

When the flags FPO BLOKE and _J indicate that there is no need to suppress low-frequency noise and at least one of FBG BLOKE and BLAKE J indicates that the reduction of blocking artifacts is not necessary, or when at least one of the first flags FPO BLOKE and _J specifies whether to suppress low-frequency noise, and the difference between two adjacent pixels in the horizontal boundaries of the unit does not exceed the quantization index N.263 (QP), is the second filtering (weak filtering release) to change the values of the pixels whose number is smaller than the first filter.

In addition, filtering the pixels at the vertical boundaries of the unit is the same as for the pixels at the horizontal boundaries of the block, using FBW.

At the same time, if the frame, kolorirovaniya to reduce blocking artifacts, when FP BLOKE indicates that there is no need to suppress low-frequency noise, one of the blocks: BLOCK and _J is internal, and FBG both units - BLOKA and _J - point to the need to reduce blocking artifacts, the first filter (strong filtering release) to change the number of values of pixels in the horizontal boundaries of the block between BLOXOM and _J.

If FP BLOKE indicates that there is no need to suppress low-frequency noise, one of the blocks (BLOCK and _J) is the internal (INTRA), any of FBG BLOKE and _J indicates that there is no need to reduce the blocking artifacts, and the difference between two adjacent pixels in the horizontal boundaries of the unit does not exceed the quantization index N.263 (QP), is the second filtering (weak filtering release) for value changes, the number of which is smaller than the first filter.

When FP BLOCK 1 indicates that there is no need to suppress low-frequency noise, no BLOCK or _J is not internal (INTRA), one of the flags FPO BLOKE and _J is in the state "1" and FBG both units: BLOKE I and _J are in a state of "1" is the first iodavleniya low-frequency noise, no BLOCK or _J are not internal, one of the first flags AF BLOKE and _J is in the state "1", and at least one of FBG BLOKE and _J is in the state "0", is the second filtering (weak filtering release).

If FP BLOKE I is not equal to zero, is the second filtering (weak filtering release).

In addition, filtering the pixels at the vertical boundaries of the unit is the same as for the pixels at the horizontal boundaries of the block, using FBG.

In addition, the first and the second filter are the same as in the first embodiment.

2.2. The angular compensator release

Compensation of the angular emission is performed only for the internal frame. The angular emission is characterized by the presence of a pixel that is either significantly more or significantly less of neighboring pixels at a corner point of the 8x8 block, as shown in Fig. 6A-6C. In Fig. 6A, where the dark gray area is divided into four blocks and one or two pixel dark gray area is located at corner points of the neighbouring blocks, with a corner point can be distorted by quantization of the DCT coefficients, as shown in Fig. 6B. This distortion of the angular points of the n frequencies. To reduce the angular emission of such emission should be detected and appropriately compensated. In Fig. 6C shows a simple coordination of its discovery, where A, B, C and D of the pixel values of the corner points of the block of 8x8 pixels.

The algorithm for detecting the angular emission is expressed as follows:

value [0] = A; value [1] = B;

value [2] = C; value [3] = D;

Average = (A+B+C+D+2)/4

Score = 0;

for (m=0; m < 4; m++)

if (|size [m] - Mean|) > QP)

The score ++; /*number of points-candidates*/,

where QP is the quantization index N.263, "Account" is a variable for storing the number of possible pixel angular emission. It is assumed that A1and A2the pixels adjacent to the pixel A, and A3-the pixel located diagonally with respect to A pixel in the same block of pixels A. If the score is zero, the angular emission is absent. If A is the only possible point in Fig. 6C and |A-A3| does not exceed 3QP12, the compensation of the angular emission is performed for A, A1and A2in the following way. Suppose that the value of compensation for A, A1and A2- A', A'1and A'2accordingly, A, A'1and A'2defined by equation (1).

A' = (4A+B+C+2D+4)/8

A'1= (A'+2A3+B3+C3+D3+2)/4, and compensation of angular emission is performed at this point, as in the case of only one candidate. Here B3WITH3and D3indicate pixels that are diagonal with respect to the pixels A, B, C respectively.

2.3. A filter for reducing image noise (low frequency noise)

As in the first variant of realization, before applying a filter to suppress low-frequency noise, investigated the interference flag. If the interference flag of the current block is in the state "1", the current block is filtered to suppress low-frequency noise. In order to protect parts of an image from being distorted when filtering, before this operation is a simple edge detection image. As shown in Fig. 7A and 7B, edge detection and two-dimensional adaptive filtering of the signal applied to the 8x8 block with the flag of interference is not equal to zero. The fact that the edge pixels are smoothed out by the filter release.

As in the second embodiment, before applying the filter for each block, analyzed flags interference, that is, FPO and FP. The algorithm to determine whether to perform the filtering for suppression of low-frequency noise is expressed as follows.

If the internal frame is - does the flag of interference BLOKE filter to suppress low-frequency noise. Filter to suppress low-frequency noise is performed, if such filtering is required; otherwise, filtering to suppress low-frequency noise is not running.

In the case of the intermediate frame, the decision to perform filtering to suppress low-frequency noise is performed. If BLOCK I is not the inner frame and is in the prediction mode of the block 8x8, then filter to suppress low-frequency noise. If BLOCK I is an internal block and is not in the prediction mode of the block is 8x8, and the flag FP BLOKE indicates the need for filtering low frequency noise, this filtering is performed. If the flag FP BLOKE indicates no filtering low frequency noise, the filtering is not performed.

Filter to suppress low-frequency noise includes edge detection image and two-dimensional adaptive filtering of the signal. To prevent distortion of image detail when filtering before this operation is a simple edge detection. Edge detection and two-dimensional adaptive filtering of the signal applied to the 8x8 block, if the above condition is satisfied, the suppression of low-frequency noise, as shown in Fig. 7A and 7B.

The following procedure describes how edge detection image in the first and second embodiments, filter to suppress low-frequency noise. The operators of one-dimensional horizontal and vertical gradient are applied to the restored blocks to Ocala quantization N. 263 QP. In order to apply the two-dimensional adaptive filtering of the signal to 8x8 pixels, it is necessary to obtain information about the edges of the image for a block of 10 x 10, which is the current block, as shown in Fig. 7B. When detecting horizontal edges, provided that the pixel [m, n] is the current pixel, a pixel [m] [n+1] is located to the right of the pixel [m, n] and the pixel [m] [n-1] is located to the left of the pixel [m, n], the difference between the pixel [m][n] and the pixel [m] [n+1] is A1, the difference between the pixel [m][n] and the pixel [m][n-1] is A2; QP - quantization index N.263 when a condition is met ((A1>QP) and (A2>QP)) or (A1>2QP) or (A2>2QP), the current pixel is detected as an edge pixel and the display edge, the edge [m][n], becomes equal to "1". Upon detection of vertical edges, provided that the current pixel is pixel [m, n], pixel [m+1][n] is located above the pixel [m, n] , pixel [m-1][n] is located below the pixel [m, n]; the difference between the pixel [m][n] and the pixel [m+1][n] = A'1, the difference between the pixel [m] [n] and the pixel [m-1][n] = A'2, while QP is the quantization index N.263 when a condition is met ((A'1>QP) and (A'2>QP)) or (A'1>2QP) or (A'2>2QP), the current pixel is detected as an edge pixel, and the display edge edge [m][n], becomes equal to "1".

/*Detect horizontal edges*/

A1 = |pixel [m][n]-pixel [m][n+1]|;

A2 = |pixel [m][n]-pixel [m][n-1]|;

if (((A1 > QP) and (A2>QP)) or (A1>2QP) or (A2>2QP))

Edge [m][n] = 1;

otherwise, {/*Detect vertical edges*/

A1 = |pixel [m][n]-pixel [m+1][n]|

A2 = |pixel [m][n]-pixel [m-1][n]|;

if (((A1 > QP) and (A2>QP)) or (A1>2QP) or (A2>2QP))

Edge[m][n] = 1;

The following describes the filtering for suppression of low-frequency noise using a two-dimensional adaptive lowpass filter. Filtering for smoothing noise involves the suppression of noise without noticeable loss of image detail. Filter to suppress low-frequency noise in accordance with the present invention is a simple convolution operation, in which the weighting coefficients for the convolution changed in accordance with the regional image. Adaptive signal filtering is applied to the decoded block using contour image [m] [n]. Kernel for two-dimensional adaptive filtering of the signal shown in Fig. 7A. When the Central point A of the filter box in Fig. 7B is located on the edge of the pixel, the operation of the two-dimensional filtering is not performed (case 1 in Fig. 7B). If no point of the edge image is that boundary point, not located in the Central point, are 4-element filter window is weighted filtering to exclude the edge pixels (Example 3 and Fig. 7B). The weighting coefficients are determined, taking into account the complexity of the calculations, so that the adaptive signal filtering can be performed by a simple shift with additional operations, as shown in the Table

3. Filling flags lock

Filling flags lock is performed in the second embodiment. In the case of QCIP, Bf_ Y (kill bits for Y) consists of a matrix h and both flags: Bf_ Cb (lockout flag for Cb) and Bf_Cr (flag lock for Cr), consist of a matrix h, where each element contains information about the noise of each block. However, when using unrestricted motion vectors, the vectors of displacement may be at a point outside of the image. Therefore, is filling Bf_Y, Bf_Cb and Cr_Bf to cover the motion vectors outside the image. If Bf_Y upper/lower elements of the column Bf_Y copied vertically to extend Bf_Y, and is horizontal reseeding in the elements of the left/right column Bf_Y after repeated vertical padding. Finally, assuming that more is whether the original size BfCb and Cr_ M/2N/2, the resulting size is equal to (M/2+4)(N/2+4). In Fig. 8 shows the result of filling Bf_Y. The filling can be performed using various methods. For example, the filling may be performed by a copy of the locking flag keyframe. When you enter a new frame, which must be subjected to the loop-filter, the filling is done only once. Then get the kill bits corresponding macroblock using filled the blocking flags.

The invention can be used in digital computing General purpose machines that perform the computing environment, including, without limitation, storage media such as magnetic media (in particular, ROM, floppy disks, hard disks and so on), optically readable media (CD-ROM, DVD and so on) and carrier waves (e.g., transmission in the network "Internet"). Therefore, the present invention can be implemented in a computer environment by using the read program code for loop-filtering and reduce the effect of quantization for encoding and decoding image data, the tool reads the program code in the computer environment includes a computer means of read the ation using distribution KOP inverse quantized image data, and the displacement vector indicating the difference between the previous frame and the current frame, and computer means reading the program code to filter the image, respectively, flag predefined way, if the selected flag specifies whether the loop-filtering the data.

Functional program, code and code segments used to implement the present invention can be developed by a qualified programmer on the basis of the above description.

When restoring the image data with high compression occurs adverse effect of quantization, such as blocking artifacts, the angular emission and low-frequency noise transients. The way the loop filter according to the present invention reduces the quantization noise due to the use of flags and adaptive filter. The blocking flags and flags of interference of each unit to help reduce the amount of computations in the loop-filter. To highlight flags lock and flags interference of the current block is the displacement vector of the intermediate frame.

When coding to facilitate obtaining visocekas what Britanie, you need to consider the computational complexity and signal-to-noise. The method according to the present invention can be performed by parallel processing without multiplication, and division, thus reducing the complexity of the hardware.

The loop-filtering according to the present invention can significantly improve subjective quality, keeping all the details of the image, so that the loop-filtering can be found the widest application.

1. The way the loop-filtering to reduce the effects of quantization that occurs when encoding and decoding the image data, characterized in that the allocate flag that indicates whether the data require image loop-filtering using the distribution coefficients of the inverse transform (DCT) is inversely converted image data and the displacement vector indicating the difference between the previous frame and the current frame, and if the flag indicates whether to perform the loop-filter, the filter corresponding to the flag data image pre-defined way that ensures the allocation of the bit-stream interference flag and the flag of the lock.

2. The method according to p. 1, characterized in that the said flag fidalgo loop filter is an internal frame, and separated from the residual signal and the reference frame, if the frame to be loop-filtered is an intermediate frame.

3. The method according to p. 2, characterized in that the said flag contains the lockout flag indicating whether the reduction of blocking artifacts at block boundaries, and the interference flag that indicates whether the reduction of low-frequency noise at the edges of the image.

4. The method according to p. 3, wherein when the image data represent the inner frame, the lock flag of the inner frame contains the lockout flag horizontally (FBG) and the lockout flag vertically (FBW), with a selection of FBG and FBW inner frame includes: definition of the DCT coefficients is inversely transformed block 8 x 8, after the compressed image data were subjected to reverse transformation, assuming that a pixel corresponding to the coefficient of the DC component, is located in the upper far left corner block of 8 x 8, consisting of 64 pixels, pixel B is located to the right of the pixel A and pixel C is located below the pixel A, the installation FBG and FBW in the state "1", indicating the need to implement the CEC is the need to perform loop filtering, if only the top row is inversely transformed block 8 x 8 is not equal to zero, and the installation FBG in the state of "1" indicating the need to perform loop filtering, if only the far left column is inversely transformed block 8 x 8 is not equal to zero.

5. The method according to p. 3, characterized in that the allocation interference flag (AF) of the inner frame, assuming that a pixel corresponding to the coefficient of the DC component is at the top in the far left corner block of 8 x 8, consisting of 64 pixels, the pixel B is located to the right of the pixel A and pixel C is located below the pixel A, AF set to "1" state, indicating the need to perform loop filtering when any pixel, in addition to pixels A, B, and C back transformed data block 8 x 8, has a coefficient not equal to zero.

6. The method according to p. 3, characterized in that the lock flag of the current intermediate frame contains FBG and FBW, while assuming that the reference frame includes a predetermined number of reference blocks, the block of the reference frame, the predicted displacement vector (MVx, MVy) block Ac current intermediate frame is the unit of displacement X, and the selection FBG and FBW in block Acthe current split timing the logical operations "And" flags FBG and FBW thrust blocks, having a predetermined number of overlapping pixels, and installation of the bitwise logical operators "And", in the form of FBG and FBW block Acthe current plane video object (PVO).

7. The method according to p. 6, characterized in that the selection FBG and FBW in block Acthe current intermediate frame further comprises the step of setting the kill bit reference macroblock as a flag block of the current macroblock, if the displacement vector of the current macroblock is a skip macroblock having zero displacement vector.

8. The method according to p. 7, characterized in that the selection FBG and FBW in block Acthe current intermediate frame further includes a filling block flag after duplication block flag.

9. The method according to p. 6, characterized in that the predetermined maximum number of overlapped pixels is 2 x 2 pixels.

10. The method according to p. 8, characterized in that the predetermined maximum number of overlapped pixels is 1 x 1 pixels.

11. The method according to p. 3, characterized in that the selection flag of interference in the block Acthe current intermediate frame, if the frame image is an intermediate kalaywa of the coefficients of the inverse conversion (CPC) of the residual signal of the intermediate blocks of 8 x 8 is not equal to zero, and installation of interference flag to "0" if all variables in the CPC are zero, and installation of FC in the state "1" if the current block is a prediction mode of the block 8 x 8 for transmitting the displacement vector per macroblock.

12. The method according to p. 3, wherein, if the image data represent an intermediate frame, the flag of the interference of the intermediate frame includes the first flag interference (FPO) and the second flag interference (FP), and the selection of the first and second flags interference FPO and FP in block Acthe current intermediate frame includes the stages of: setting a first flag interference FPO current block Ac in state "1", if the constant components of the coefficients of the inverse conversion (CPC) of the residual signal of the intermediate macroblocks 8 x 8 is not equal to zero, otherwise, the installation FPO current block Acto "0", and setting the second flag interference FP of the current block is in state "1" if any variable component of the CPC of the residual signal of the intermediate frame 8 x 8 is not equal to zero; otherwise, the installation FP of the current block is Acto "0".

13. The method according to p. 3, wherein, provided that a predetermined block constituting a frame to be loop-filter, antohny the previous unit I does not exceed the predetermined value and the difference between the current block J, and the quantized previous unit J does not exceed the predetermined value, filtering the release stage filtering is not performed.

14. The method according to p. 4, characterized in that, provided that a predetermined block constituting a frame to be loop-filtering is a block I and block adjacent to the block I is a block J, filter release to reduce blocking artifacts, which includes the steps of: performing a first filtering for changing a predetermined number of values of pixels in the horizontal boundary block between blocks I and J, if the flags interference blocks I and J indicate the absence of suppression of low-frequency noise, and both FBG flag blocks I and J indicate the need for reduction of blocking artifacts, and comparing the difference between two adjacent pixels on the border of the block with the index of the quantization QP standard H. 263, if both flags interference blocks I and J indicate the absence of suppression of low-frequency noise and at least one of FBG blocks I and J indicates no need for reducing blocking artifacts, or if at least one of the flags interference blocks I and J indicate the need to suppress low-frequency noise, and change prevailed two pixels less than the quantization QP.

15. The method according to p. 5, wherein provided that a predetermined block constituting a frame to be loop-filtered is a block I and block adjacent to the block I is a block J, perform the filtering for reducing blocking artifacts, comprising the steps: performing a first filtering for changing a predetermined number of values of pixels in the horizontal boundary block between blocks I and J, if the flags interference blocks I and J indicate the absence of suppression of low-frequency noise, and both FBG flag blocks I and J indicate the need for reduction of blocking artifacts, and comparing the difference between two adjacent pixels on the border of the block with the index of the quantization QP standard H. 263, if both flags interference blocks I and J indicate the absence of suppression of low-frequency noise, and at least one of FBG blocks I and J indicates no need for reducing blocking artifacts, or if at least one of the flags interference blocks I and J indicate the need to suppress low-frequency noise, and changes a predetermined number of values of pixels, which is less than this number for the first filtrate fact, that provided that a predetermined block constituting a frame to be loop-filtered is a block I and block adjacent to the block I is a block J, filter release to reduce blocking artifacts, comprising the steps: performing a first filtering for changing a predetermined number of values of pixels near the border of the block between blocks I and J, if the interference flag of the block I indicates no need for suppression of low-frequency noise and both units I and J are internal blocks, and comparing the difference between two adjacent pixels on the border of the block with the index of the quantization QP standard H. 263, if the interference flag of the block I indicates no need for reducing low-frequency noise and at least one of FBG blocks I and J indicates no need for reducing the blocking artifacts, and performing a second filtering for changing a predetermined number of values of pixels that is less than this number for the first filter if the difference between two pixels is less than the quantization QP of the first filter, if the flag is interference of the block I indicates no need for reducing nyskohus the th coefficient of the inverse conversion (CPC) of the residual signal of the block I and J, and both FBG blocks I and J represent "1"; the second filter if the flag interference of the block I indicates no need for reducing low-frequency noise, with no block I or block J are inner blocks, there is a constant component of the coefficient of the inverse conversion (CPC) of the residual signal of the block I or J and at least one FBG blocks I and J is equal to zero, and the second filter if the flag interference of the block I indicates the need to reduce low-frequency noise.

17. The method according to p. 11, wherein assuming that a predetermined block constituting a frame to be loop-filtered is a block I and block adjacent to the block I is a block J, filter release to reduce blocking artifacts, comprising the steps: performing a first filtering for changing a predetermined number of values of pixels near the border of the block between blocks I and J, if the interference flag of the block I indicates no need for reducing low-frequency noise, and both units I and J are internal blocks, and comparing the difference between two adjacent pixels on the border of the block with the index of the quantization QP hundred at least one of FBG blocks I and J indicate the absence of a necessary reduction of blocking artifacts, and performing a second filtering for changing a predetermined number of values of pixels that is less than this number for the first filter if the difference between two pixels is less than the quantization QP, the first filter if the flag interference of the block I indicates no need for reducing low-frequency noise, no block I or block J are inner blocks, there is a constant component of the coefficient of the inverse conversion (CPC) of the residual signal of the block I and J, and both FBG blocks I and J is equal to "1"; the second filter if the flag interference of the block I indicates no need for reducing low-frequency noise, no block I or block J are inner blocks, there is a constant component of the coefficient of the inverse conversion (CPC) of the residual signal of the block I or J and at least one FBG blocks I and J is equal to zero, and the second filter if the flag interference of the block I indicates the need to reduce low-frequency noise.

18. The method according to p. 7, wherein under the condition that a predetermined block, stood the block J, filter release to reduce blocking artifacts, comprising the steps: performing a first filtering for changing a predetermined number of values of pixels near the border of the block between blocks I and J, if the second flag interference FP of the block I indicates no need for reducing low-frequency noise, with unit I and unit J is an internal block and both FBG blocks I and J indicate the need for reduction of blocking artifacts, comparing the difference between two adjacent pixels on the border of the block with the index of the quantization QP standard H. 263, if FP of the block I indicates no need for reducing low-frequency noise, the block I and J is an internal block, and at least one of FBG blocks I and J indicates no need for reducing the blocking artifacts, and performing a second filtering for changing a predetermined number of values of pixels that is less than this number for the first filter if the difference between two pixels is less than the quantization QP, the first filter if the flag FP of the block I indicates no need for reducing low-frequency noise, no block I or block J is not Aleutskaya second filter, if the flag FP of the block I indicates no need for reducing low-frequency noise, no block I or block J are inner blocks, at least one of the flags FPO is an internal block, at least one of the flags FPO blocks J and P, equal to "1", and at least one of FBG blocks I and J is not equal to "1", and the second filter if the flag FP block I is not equal to zero.

19. The method according to p. 12, characterized in that, provided that a predetermined block constituting a frame to be loop-filtering is a block I and block adjacent to the block I is a block J, filter release to reduce blocking artifacts, comprising the steps: performing a first filtering for changing a predetermined number of values of pixels near the border of the block between blocks I and J, if the second flag interference FP 1 block I indicates no need for reducing low-frequency noise, block I and block J is an internal block and both FBG blocks I and J indicate the need for reduction of blocking artifacts, comparing the difference between two adjacent pixels in the block with the index of the quantization QP standard H. 263, if FP block I indicates n is ENISA least one of FBG blocks I and J indicates no need for reducing the blocking artifacts, and perform the second filtering for changing a predetermined number of values of pixels that is less than this number for the first filter if the difference between two pixels is less than the quantization QP, the first filter if the flag FP of the block I indicates no need for reducing low-frequency noise, no block I no block J are inner blocks and at least one of the flags FPO blocks I and J is equal to "1", and both FBG blocks I and J is equal to "1", the second filter if the flag FP of the block I indicates no need for reducing low-frequency noise, no block I no block J are inner blocks, at least one of the flags FPO blocks I and J is equal to "1" and at least one of FBG blocks I and J is not equal to "1", and the second filter if the flag FP block I is not equal to zero.

20. The method according to p. 19, characterized in that provided that six pixels on the horizontal block boundary between the blocks I and J are pixels A, B, C, D, E and F, and the pixels C and D is the nearest pixels on the horizontal edge of the block, the pixels A and F - pixels, the most remote of the 1,1,1) for the six pixels in the processes specified by the first filter, the second filtering is performed on the pixels C and D; in which two pixels C and D is replaced by the expression C=C+(D-C)/4 and D=D+(D-C)/4, if the absolute value of the difference between pixels C and D is less than the quantization QP standard H. 263.

21. The method according to p. 3, characterized in that the filter to reduce low-frequency noise on stage filtration perform, if the interference flag specifies whether filtering to reduce noise, and do not perform if the interference flag indicates no filtering for noise reduction and filtering to reduce low-frequency noise includes the steps of: (a) detecting horizontal and vertical edges of the image data and (b) performing a two-dimensional adaptive filtering of the signal for a block of 8 x 8 in which it is necessary to reduce the low-frequency noise, while at stage (a) for edge detection in the horizontal direction under the condition that the pixel [m,n] is the current pixel, a pixel [m][n+1] is located to the right of the pixel [m,n] and the pixel [m][n-1] is located to the left of the pixel [m,n], the difference between the pixel [m][n] and the pixel [m] [n+1] is A1, the difference between the pixel [m][n] and the pixel [m][n-1] is A1 and QP is the quantization index of the standard H. 263, when a condition is met ((A1 > QP) and (A1 > QP)) or (A1 > 2QP) or (A2 > 2QP), those who(a) for edge detection vertical assuming that the current pixel is pixel [m,n], pixel [m+1][n] is located above the pixel [m,n], the closer the pixel is [m-1][n], the difference between the pixel [m][n] and the pixel [m+1][n] is A'1, the difference between the pixel [m] [n] and the pixel [m-1][n] is A'2, and QP is the quantization index of the standard H. 263, when a condition is met ((A'1 > QP) and (A'2 > QP)) or (A'1 > 2QP) or (A'2 > 2QP), the current pixel detect as edge pixel, and the display region, the edge [m][n], becomes equal to "1", and step (b) for the adaptive filtering of the signal 4-element filter window is applied to the block 8 x 8 to determine whether the Central pixel of the filter window edge pixel, the filtering is performed if the Central pixel is an edge pixel, otherwise performing a weighted filtering.

22. The method according to p. 3, wherein, provided that a predetermined block constituting a frame to be loop-filtered is a block I and block adjacent to the block I is a block J, filter to reduce low-frequency noise on the stage filtering is not performed if the difference between the current block I and the previous quantized block I is not bigger than a pre-determined specific value.

23. The method according to p. 5, wherein provided that a predetermined block constituting a frame to be loop-filtered is a block I and block adjacent to the block I is a block J, filter to reduce low-frequency noise on stage filtering performed if the flag interference of the block I indicates the necessity of performing loop-filtering on the phase of the filter, otherwise the filter to reduce low-frequency noise does not perform.

24. The method according to p. 12, characterized in that the filter to reduce low-frequency noise perform, if block I is an internal block, if block I is an internal block and is forecast 8 x 8, and if the block I is an internal block and is not in the prediction mode 8 x 8 and the flag FP of the block I indicates no need for the filter to reduce low-frequency noise, the specified filtering is not performed if the flag FP of the block I indicates no need for the filter to reduce low frequency noise.

25. The method according to p. 22, characterized in that the filtering to reduce low-frequency noise includes the steps of: (a) detecting horizontal and vertical edges of the data izobratetalny noise, at stage (a) to detect horizontal edges, provided that the pixel [m,n] is the current pixel, a pixel [m][n+1] is located to the right of the pixel [m,n] and the pixel [m][n-1] is located to the left of the pixel [m,n], the difference between the pixel [m][n] and the pixel [m][n+1] is A1, the difference between the pixel [m] [n] and the pixel [m][n-1] is A2 and QP is the quantization index of the standard H. 263, when a condition is met ((A1 > QP) and (A2 > QP)) or (A1 > 2QP) or (A2 > 2QP), the current pixel detect as edge pixel, and the display region, the edge [m][n], becomes equal to "1", and step (a) to detect vertical edges, provided that the current pixel is pixel [m,n], pixel [m+1][n] is located above the pixel [m,n], pixel [m-1][n] is located below the pixel [m,n], the difference between the pixel [m][n] and the pixel [m+1][n] is A'1, the difference between the pixel [m][n] and the pixel [m-1][n] is A'2, and QP is the quantization index of the standard H. 263, when a condition is met (((A1 > QP) and (A2 > QP) or (A1 > 2QP) or (A2 > 2QP), the current pixel detect, as the edge pixel, and the display region, the edge [m][n], becomes equal to "1", and step (b) for the adaptive filtering of the signal 4-element filter window is applied to the block 8 x 8 to determine whether the Central pixel of the filter window is the case of performing a weighted filtering.

26. The method according to p. 23, wherein filtering to reduce low-frequency noise includes the steps of: (a) detecting horizontal and vertical edges of the image data and (b) performing a two-dimensional adaptive filtering in block 8 x 8, which must be removed low-frequency noise, while at stage (a) to detect horizontal edges, provided that the pixel [m, n] is the current pixel, a pixel [m][n+1] is located to the right of the pixel [m, n] and the pixel [m][n-1] is located to the left of the pixel [m,n], the difference between the pixel [m] [n] and the pixel [m][n+1] is A1, the difference between the pixel [m] [n] and the pixel [m][n-1] is A2 and QP is the quantization index of the standard H. 263, when a condition is met ((A1 > QP) and (A2 > QP)) or (A1 > 2QP) or (A2 > 2QP), the current pixel detect as edge pixel, and the display region, the edge [m][n], becomes equal to "1", and step (a) to detect vertical edges, provided that the current pixel is pixel [m,n], pixel [m+1][n] is located above the pixel [m,n], pixel [m-1][n] is located below the pixel [m,n], the difference between the pixel [m][n] and the pixel [m+1][n] is A'1, the difference between the pixel [m][n] and the pixel [m-1][n] is A'2, and QP is the quantization index of the standard H. 263, satisfied the condition ((A'1 > QP) and (A'2 > QP)) which becomes equal to "1", and phase (b) for the adaptive filtering of the signal 4-element filter window is applied to the block 8 x 8 to determine whether the Central pixel of the filter window edge pixel, and the filtering is performed if the Central pixel is an edge pixel, otherwise performing a weighted filtering.

27. The method according to p. 1, characterized in that for reducing the angular emission generated at the point where the corners of four blocks, when encoded and decoded block image data, further includes the steps of: (a) detection of the angle of ejection of the block 8 x 8 back the converted image data and (b) compensation of the detected angle of emission, and phase (a) detection, provided that A, B, C, and D are four pixels around the point where the four corners, and value [0], value [1], value [2] value [3] - pixel values A, B, C and D, respectively, the average values of the pixels are set as (A+B+C+D+2)/4, the difference between the value of each pixel in average compared to the index of the quantization QP standard H. 263 and the number of candidate pixels on the angle of emission accumulate to detect a specified angle of emission, if the difference exceeds the index of the quantization QP, and atapy with respect to the pixel A, in the same block of pixel A and A', A'1and A'2- compensated pixel values A, A'1and A'2accordingly, if there is only one candidate angle of emission, the detected candidate is a pixel and A difference between pixel A and A3less than 3QP/2, the compensation of the angular emission is carried out using equations

A'=(4A+B+C+2D+4)/8;

A'1=(A'+3A1+2)/4;

A'2=(A'+3A2+2)/4,

and, if the number of candidates is greater than two, choose the candidate that has the largest difference from the expression (A3+B3+C3+D3+2)/4, and the compensation of the angle of emission is performed at this point as well, as in the case of only one candidate.

28. Device for loop-filtering the image data, mitigating the effect of quantization that occurs when encoding and decoding the image data, characterized in that it contains block allocation flag allocation flag that indicates whether the loop-filtering the image data using the distribution coefficients of the inverse conversion (CPC) of the image data and the displacement vector indicating the difference between the previous frame and the current frame, the filter deblocare is adelene flag the compensator angle of emission for the detection of the angular emission is inversely converted and filtered data and to compensate for the detected angle of emission and the filter to perform filtering with reduced low-frequency noise data with compensation of the angular emission by analyzing the interference flag, the selected block selection flag, and these flags include flag lock, indicating the need to reduce blocking artifacts at block boundaries, and the interference flag indicating the need to reduce low-frequency noise at the edges of the image.

29. The device according to p. 28, characterized in that the selection flag of the block selection flag to perform internal flag and intermediate flag when allocating internal flag used the distribution of the coefficients of the inverse conversion (CPC) is inversely converted image data, and in the interim allocation flag using the displacement vector indicating the difference between the previous frame and the current frame.

30. The device according to p. 28, characterized in that the filter release is a one-dimensional horizontal and vertical lowpass filter.

31. The device according to p. 28, characterized t is the priority points:

01.08.1998 - PP.1 - 6, 11, 13 - 17, 19 - 23, 25 - 31;

02.11.1998 - PP.7 - 10, 12, 18, 24.

 

Same patents:

FIELD: re-synchronization.

SUBSTANCE: method can be used in decoding channel according to MPEG-4 standard. To provide proper decoding of pressed video data signal, the re-synchronization word RW differs from known words of variable length code VLC as well as from start code of plane of video object and has at least 17 sequent zeros, after which the unit follows, for plane of video object coded to provide two-directional prediction. After error in transmission in pressed video signal is detected, the pressed video data signal can be re-synchronized.

EFFECT: higher efficiency of re-synchronization.

4 cl, 2 dwg

FIELD: image transferring equipment engineering, possible use in multimedia communications.

SUBSTANCE: in accordance to method, when error codes are detected on receiving side, data of code stream of image with error codes are refused prior to decoding of data of code stream of image, and refused data of code stream of image are replaced with data of code stream of image, positioned in appropriate position of previous frame, and data of code stream of image are encoded continuously. Also, an array of marks is set up for data of code stream of image prior to encoding on receiving side, to perform recording of positions, where error codes have been detected.

EFFECT: possible avoidance of transfer of internal frame images on transmitting side and of frozen images on receiving side, or decrease of their occurrence periods, thus improving quality of image.

7 cl, 2 dwg

FIELD: information technology.

SUBSTANCE: invention relates to buffering packets of a media stream during transmission from a transmission device to a receiving device. Media packets are generated from at least one type of media information in a stream generator; at least one transmission frame is generated based on transmitted media packets; transmitted packets are generated from at least one transmission frame and a transmission schedule is generated for transmitted packets. In addition, the first and second steps of hypothetical decoding are executed. The first step of hypothetical decoding is executed in accordance with the transmission schedule, and involves buffering the transmitted packets in accordance with the transmission schedule in the first buffer for hypothetical decoding and output of packets from the first buffer for hypothetical decoding based on the transmission frame. The second step of hypothetical decoding involves controlling occupance rate of the first and second buffer for hypothetical decoding by controlling at least one of the following: operation of the stream generator, generation of at least one transmission frame, transmission schedule.

EFFECT: more efficient buffering of media stream packets.

20 cl, 7 dwg

FIELD: information technologies.

SUBSTANCE: method and device are suggested for multilevel integration used for elimination of errors. Error is detected in multimedia data on the basis of the first level protocol, and then error detected in multimedia data is masked on the basis of the second level protocol. In one aspect error in multimedia data is eliminated on the basis of communication level protocol, and it is controlled on the basis of transport level protocol. Further distribution of controlled error is determined on the basis of synchronisation level protocol, then error detected in multimedia data is masked on the basis of applied level protocol. Further stage of error elimination and scaling stage are provided.

EFFECT: increased efficiency of multimedia data stream processing by reception of multiple streams of coded multimedia data, eliminating errors in erroneous part of stream and recovering multimedia data from multiple streams.

40 cl, 10 dwg

FIELD: physics; image processing.

SUBSTANCE: invention relates to a method of buffering multimedia information, as well as a method of decoding a coded stream of images in a decoder, in which the coded stream of images is received in form of transmission blocks which contain multimedia data. A system for processing multimedia data is proposed, which contains a coder for coding images and a buffer for buffering multimedia data. Multimedia data are included in the data transmission blocks. The data transmission blocks are ordered in the transmission sequence, which at least partially differs from the sequence of decoding multimedia data in transmission blocks. There is also definition block, which can set a parametre which indicates the maximum number of data transmission blocks which precede any data transmission block in a stream of packets in the transmission sequence and that data transmission block is tracked in the decoding sequence.

EFFECT: more efficient compression when buffering multimedia information.

32 cl, 7 dwg

FIELD: physics, communications.

SUBSTANCE: invention relates to transmission of a media stream over an error-prone digital video broadcasting - handheld (DVB-H) channel in which media datagrams are labelled according to a priority, packed in a multiprotocol encapsulation section, unequally protected using forward error-correction codes packed into a traffic stream and transmitted into the channel using packets with time-division. A system and a method are proposed for transmitting a multiplexed service stream over a DVB-H channel. Media IP packets are priority labelled. For each packet with time division, the IP packets are grouped based upon the priority labels. Multi protocol encapsulation - forward error correction (MPE-FEC) matrices are made for different priority labels in each packet with time division. Reed-Solomon code data table (RSDT) columns are computed such that the average service bit rate does not overshoot the maximum allowed bit rate, and protection increases with priority. The application data table (ADT) and RSDT of the MPE-FEC matrices are then encapsulated into MPE-FEC sections.

EFFECT: shorter start delay during reception of an unequally protected priority service bit stream.

21 cl, 10 dwg

FIELD: physics, communications.

SUBSTANCE: invention relates to multimedia transmission systems, specifically to methods and a device for acquiring services. Proposed is a service acquisition device which has a source coder configured to generate one or more channel switch video (CSV) signals, which is an independently decoded version of a low-resolution video for the selected channel in a received multiplex transmission and associated one or more multimedia signals, an error coder configured to code CSV signals and multimedia signals for formation of coded error blocks, and a linker configured to encapsulate coded error blocks into a multiplex transmission signal.

EFFECT: fast acquisition of a service and/or switching between services in multiplex transmission.

60 cl, 23 dwg

FIELD: information technologies.

SUBSTANCE: video data is coded, packet is formed with coded video data, and packet is transferred via wireless channel into access network. Level of access control to transfer medium (MAC) receives negative notice from the access network (NAK). It is identified whether received NAK is associated with packet, which contains video data. If received NAK is associated with packet, which contains video data, errors are corrected.

EFFECT: improved efficiency of video data errors correction.

36 cl, 5 dwg

FIELD: information technologies.

SUBSTANCE: method for transmission/reception of signal and device for transmission/reception of signal. Device for transmission of signal includes coder with forward error correction (FEC), which executes FEC-coding of input data for detection and correction of data errors, interleaver, which interleaves FEC-coded data, and unit of symbols display, which displays interleaved data to data of symbol according to method of transmission.

EFFECT: improved efficiency of channel bandwidth use, increased speed of data transmission and increased distance of signal transmission, reduced cost of network development for signal transmission-reception.

15 cl, 33 dwg

FIELD: information technologies.

SUBSTANCE: several various VLC-tables are stored in coding devices, in process of coding and decoding, one of VLC-tables is selected and used to do coding of SVR for this video unit. Table may be selected on the basis of number of neighbouring video units for current video unit, which include non-zero transformation ratios.

EFFECT: increased coefficients of coding of SVR video units, in which structures of coefficients occurring with higher probability, are coded with the help of shorter codes, while structures of coefficients, which occur with lower probability, are coded with the help of longer codes, which is especially useful in coding of video units of improved layers in coding of scalable video.

25 cl, 7 dwg, 1 tbl

Up!