The method of image processing

 

(57) Abstract:

The invention relates to computing. Its use in the methods of image processing can improve the sharpness and contrast of the processed image while suppressing noise. The method is based on the initial image in the form of a matrix of picture elements, dividing the original image into n frequency channels, each of which contains a matrix of the same dimension as the original image, the edge detection image and generating the output image of the preceding n frequency channels based on the selected boundaries. The technical result is achieved due to the fact that the separation of the image into frequency channels emit low-frequency channel and n-1 high frequency channels, for marking boundaries in high-frequency channels for each picture element calculate the correlation between it and the surrounding image elements and compares the correlation value with the correlation values for the respective position of the image elements in the other frequency channels and the threshold for this channel, by comparing the results formed the considerations applying summarize the picture elements of the low-frequency channel with all the relevant position of the image elements of the n-1 high frequency channels, multiplied by corresponding weighting coefficients. 22 C.p. f-crystals, 8 ill.

The invention relates to automatic control and computer engineering, in particular to methods of pre-processing of the images to make them more sharpness and contrast.

There is a method of image processing according to the patent of Russian Federation N 2015561 from 16.04.91, MKI G 06 K 9/40, in which image correction is carried out on the basis of the analysis of the original image at the current point and the local mean of the signal in it some neighborhood.

The original image is smooth, i.e., to create an image that contains only the low frequency component corresponding to the background. Then the smoothed image is subtracted from the source, i.e. the source image subtract the background component. The result is an image that contains only high-frequency components without the background, which then reinforce and summarize with the smoothed image.

The disadvantage of the above method is that reinforce not only the signal but also noise, which contain high frequency components, which reduces the quality of the output image.

A method of processing electronic and obtained the smoothed image is subtracted from the source, receiving an image that contains only high-frequency component. The resulting image is subjected to adaptive gain at which the weighting factor is greater, the larger the difference values of neighboring picture elements from the Central. The output image is formed by adding the adaptive enhanced high-frequency components and the original image, the result is a sharpening without amplifying noise.

The technical solution described in the above patent does not provide noise suppression, as the proposed method can only enhance high-frequency components of the image containing noise.

In addition, the disadvantage described above is similar to the impossibility of providing a significant sharpening weakly visible parts of the image, since such increase requires strengthening the high-frequency component in the image where the difference between adjacent elements from the Central comparable with the noise, therefore, the enhancement of high-frequency component in such areas leads to increased noise.

There is a method of image processing according to the U.S. patent N 5381490 published 10.01.95, MKI G 06 K 9/40, which produce vychisleniya obtained difference choose one of three modes:

- strengthening of borders by enhancing high frequency components, if T1where T1the first predetermined threshold value;

- transmission of the original image unchanged if T2< < T1where T2the second predefined threshold value;

- smoothing of the image (noise reduction), if < T2.

One of the drawbacks of the above-described method is that the choice of mode of work carried out by comparing the greatest difference with the threshold values, may lead, in case of change of this difference near one of the threshold values for several neighboring image elements, to the electoral increased noise caused by the different processing modes of the neighboring picture elements.

In addition, the threshold value T1and T2not evaluated adaptive, which leads to the impossibility of qualitative processing of images with different noise levels without adjustment.

Another approach to the problem of noise reduction in image processing is described in U.S. patent N 5563963 published 08.10.96, MKI G 06 K 9/40, in which for each element of the original image chosen by many groups of its neighbors, the image can be anywhere in the square. For each of the selected groups of image elements using the least squares method produces an approximation of this group is a linear function of the (planar approximation), compute the error of this approximation and determine the new value of the selected item according to this approximation.

Element of the output image is formed by the weighted sum of the new values of the selected item in all groups, with the weight of the group is higher, the smaller the error of the linear approximation for this group.

The disadvantage of the above described analog is the inability of edge enhancement or image detail, since the described method produces only the aliasing noise. In addition, the above method requires high computational cost to construct linear approximations to hundreds of groups for each of the hundreds of thousands of picture elements.

The known method and device, image processing for U.S. patent N 5739922 published 14.04.98, MKI G 06 K 9/40, H 04 N 1/40, in which the original color image is divided into three isotropic frequency channel: low frequency (LF), medium frequency (MF) and high frequency (HF) channels, and then produce the adaptive RF image and ADAP who relatia between the three color components. The output image is formed by summing the LF image with adaptive weakened MF image and adaptive power RF image.

The technical solution described in the above analogy, is of limited use as it only applies to color images, because the processing of information using the correlation between the colors.

In addition, in the known analog amplify the RF components of the image containing noise, and the degree of noise suppression in the midrange components of the image is limited due to the lack of separation of the input image according to the directions. These disadvantages of the described method significantly reduces the noise.

The use of isotropic channels makes it impossible to release and strengthen the weak parts having an elongated shape, and especially the borders.

All of these issues reduce the quality of the output image.

The closest analogue is a method of image processing according to the U.S. patent N 5351305 published 27.09.94, MKI G 06 K 9/40, in which the original image by a frequency filter get lots of images filtered in the direction. The output image which is for the original image, depending on the presence or absence of a contrast border, and nearby (processed) element of the original image. The presence of a contrasting border for the selected image element is determined by calculating the eigenvectors and compare its length with a predetermined threshold.

In the absence of boundaries corresponding element of the output image is made equal to the corresponding element of the input image.

In the presence of the boundaries of the corresponding element of the output image is equal to the corresponding item is also filtered in the direction of the image in which the direction of the filter closest to the certain direction of the border.

In the above-described method of image processing when determining the boundaries of the image possible case, when the length of eigen vectors for the neighboring picture elements changes near the threshold value. In this case, it may happen selective amplification of noise caused by the sampling of neighboring image elements from different images (original and filtered in the direction), which leads to deterioration of the quality of the output image.

The selection of elements of the output image in the presence of boundaries is made only from one of the filtered in the direction of the image, which leads to a complete suppression of all features in the original image, which differ in the direction of the detected boundaries, even when these details are well visible in the source image.

Task claimed as the invention of the technical solution is to increase the sharpness and contrast of the image while suppressing noise.

The problem is solved in that in the method of processing image-based representation of the original image in the form of a matrix of picture elements, dividing the original image into n frequency channels, each of which contains a matrix of the same dimension as the original image, the edge detection image and generating the output image of the preceding n frequency channels based on the selected structures, the separation of the image on the frequency of certain channels for each picture element calculate the correlation between the selected picture element and surrounding picture elements, compares the correlation value with the correlation values for the respective position of the image elements in the other frequency channels and the threshold for this channel, by comparing the results form for each of the image elements of each of the n-1 high frequency channels weighting factor, and generating the output image is produced by summing each of the picture elements of the low-frequency channel with all relevant works on the position of the image elements of the n-1 high frequency channels and their corresponding weights.

The task is achieved by the fact that m of the n-1 high frequency channels, where 2 m n-1, differ from each other only in the direction of primary transmission. This weighting factor for each image element of each of the m high-frequency channels is determined by comparing the corresponding correlation values with a threshold value and the values of the correlations for the corresponding position of the image elements in the remaining m-1 high frequency channels.

The problem is solved also by the fact that the correlation value for each item I want the calculation of the weighted sum of neighboring elements using anisotropic weight, moreover, the direction of anisotropy corresponds to the direction of preferential transmission of the processed frequency channel.

In addition, the threshold value for each of the n-1 high frequency channels is determined by analyzing the distribution of values or absolute values of image elements of this frequency channel. It may also be that the threshold values for all frequency channels is determined by analyzing the distribution of element values of the original image or their absolute values.

Claimed as the invention has the following distinctive from the nearest similar features:

1) when dividing the image into frequency channels emit low-frequency channel and n-1 high frequency channels;

2) to highlight structures in each of the n-1 high frequency channels for each picture element calculate the correlation between the selected picture element and surrounding picture elements, compares the correlation value with the correlation values for the respective position of the image elements in the other frequency channels and the threshold for this channel, by comparing the results form the output image is produced by summing each of the picture elements of the low-frequency channel with all the relevant position of the image elements of the n-1 high frequency channels, multiplied by the corresponding weighting factors.

On the first ground, it should be noted that the selection unhandled low-frequency channel, you can ensure faithful reproduction of large parts (objects) of an image, as the objects pass through the low-frequency channel without treatment. At the same time, the energy of the noise coming through the uncultivated LFE, low, because a large part of this energy falls on the HF region.

In addition, the allocation unhandled low-frequency channel, you can speed up the calculation of correlations in the n-1 high frequency channels, so as to calculate correlations require background subtraction, i.e., the local average. This subtraction is performed by allocation (subtract) the low-frequency channel.

The selection of multiple frequency channels allows, in contrast to the selection of one or two channels, increase noise, because it does not skip the image elements of those channels in which there are no patterns about these elements of the image.

On the second distinctive feature it should be noted that the correlation between the selected element isopt.to. typical noise has a correlation value close to zero, and correlation values for several neighboring picture elements forming the structure are positive and close to each other. This, in turn, ensures a high noise reduction in combination with the strengthening of structures, i.e., contrast and sharpness of the image.

The correlation between the selected element and its neighbors to highlight structures (parts) gives a quantitative measure of the intensity structures (boundaries) in each image. This, in turn, allows the selective amplification of specific structures of power when calculating weighting coefficients, for example, weak structures. The correlation between the selected element and its neighbors also allows you to apply the described method to a wide range of images, including color, black and white and three-dimensional images.

The third distinctive feature consists in generating the output image by a weighted summation of the frequency channels, allows, in contrast to sample elements of the image output from one of the frequency channels, completely eliminate channels.

Furthermore, the definition of thresholds by analyzing the distribution of the image elements allows to provide high-quality image processing with high noise level and low without tuning parameters.

In addition, the use of anisotropic frequency channels and anisotropic scales can provide a high sensitivity to weak structures having an elongated shape, as well as to weak boundaries.

The invention is illustrated by drawings of the device that implements the method.

In Fig. 1 presents a block diagram of the device of Fig. 2-5 examples of implementing the blocks of the device of Fig. 1, where Fig. 2 - block splitting frequencies, Fig. 3 - the unit for computing the correlations in Fig. 4 - block forming weight coefficients of Fig. 5 - forming unit of the image output.

Forth in Fig. 6 shows an example of setting a predefined matrices selecting frequency channels, Fig. 7 is an illustration of the operation of the separation unit of frequency, Fig. 8 is an example of setting functional dependence of the weighting coefficients from the values of the correlations.

In accordance with drawing Fig. 1 the device comprises a sensor signal of the image 1, the output of which is Finance output image 5, the remaining outputs of the block division frequency 2 connected to respective inputs of the computing unit correlation 3 and inputs 101- 104the forming unit of the image output 5, other inputs 81- 84which are connected with the corresponding outputs of the shaping unit weights 4, the respective inputs of which are connected to respective outputs of the computing unit correlation 3, and the output processing unit of the image output 5 is connected to the input of the memory unit 6, the output of which is the output device.

In Fig. 2 presents an example implementation of block division frequency 2. In accordance with the drawing unit 2 contains a direct Fourier processor 11, an input connected to the output of block 1, and the output of the direct Fourier processor connected with the first inputs of the matrix multipliers 120- 124second inputs of which are connected to the outputs of the respective memory blocks 130- 134containing predefined matrix separation frequency channels (Fig. 6). Matrix multipliers 120- 124perform element-by-element matrix multiplication, received at their inputs, and their outputs connected to inputs of the inverse Fourier processors 140- 144. Whoopity Fourier processors 141- 144are inputs 101- 104the forming unit of the image output 5 and connected to respective inputs of the computing unit correlations 3.

In Fig. 3 presents a block calculate correlations 3 containing the memory unit 15 for storing the image of the given frequency channel, one input of which is the input of the computing unit correlations 3 and is connected to the input of the block definition noise 20 whose output is connected to one of inputs of the divider 19, a second input connected to the output of the multiplier 18, one input of which is connected to the output 21 of the memory unit 15, the other outputs 22 through which the multiplier adder 17 is connected to the second input of the multiplier 18, the second input of the memory unit 15 is connected to the block address generation for memory images 16. The output of the divider 19 is the output of the computing unit correlations 3.

The definition block noise 20 can be performed according to the scheme shown in U.S. patent N 5657401, publ. 12.08.97, M CL G 06 K 9/40.

All memory blocks are blocks of random access and executed by well-known schemes.

The multiplier adder 17 may be, for example, made in the form of eight multipliers (the number of neighbors of the chosen elements image is westwoodone inputs of the adder, the output which is the output of the unit 17. The first inputs of the multipliers are input unit 17, the second inputs of the multipliers are served predefined coefficients.

In Fig. 4 presents the set of weights 4, containing blocks of rounding 231- 234the inputs are the input processing unit weights 4, and outputs connected to inputs of the block address generation 24, the output of which is connected to the inputs of the functional converters 251- 254which are blocks of memory that stores for each set of four input values of the correlation value of the weighting factor. The functional outputs of the converters 251- 254connected with the inputs of the memory blocks 261- 264for accumulation of weights, the other inputs of these blocks is connected to the output of the block address generation 27, and the outputs of the blocks 261- 264connected to respective inputs of adder averaging the weights 281- 284the outputs are the outputs of the shaping unit weights 4.

In Fig. 5 shows the forming unit of the image output 5 containing four multiplier 291- 29<4, and second inputs 101- 104are the exits separation unit frequency 2. The outputs of the multipliers 291- 294connected to respective inputs of the adder 30, the input 9 of which is the output 7 of the block division frequency 2, and the output of the adder 30 is the input of the memory block of the output image 6.

The method is implemented by the device as follows.

Input image output from the image sensor 1 is fed to the inlet of the separation unit frequency 2 (Fig. 1). As the image sensor 1 can be used, for example, the NMR scanner, the output of which receive an image describing the slice of the object, in the form of a matrix containing discrete image elements. In the separation unit frequency 2 (Fig. 2) the input image, an example of which is shown in Fig. 7a is converted into a frequency representation of the block 11 by a direct Fourier transform. Frequency representation contains all the information that is present in the input image is a matrix of the same dimension as the input image. This matrix is fed to identical matrix multipliers 120- 124performing element-by-element matrix multiplication of the frequency representation of n is i.i.d. frequency and are in the memory 130- 134. Examples of these matrices are shown in Fig. 6.

As the image is represented as a two-dimensional matrix, its frequency representation is a two-dimensional matrix. In Fig. 6A schematically shows a matrix of frequency representation, where the axis of the spatial frequencies in the horizontal and vertical directions indicated respectively BYxand Ky.

The zero spatial frequency corresponding to a constant density (Ref. 31) image, is located at the intersection of the axes TOxand Ky.

The maximum spatial frequency in the horizontal direction is in the points POS. 32 and 33; these points are examples of input images presented in Ref. 34 and 35.

Similarly, the maximum spatial frequency in the vertical direction is in the points POS. 36 and 37; the example of the input image corresponding to these points is shown in POS. 38.

Points POS. 39 - 42 are the maximum possible values of the spatial frequency. A corresponding example of the input image is presented at POS. 43.

The average spatial frequency in the horizontal direction is point rastenij frequencies in Fig. 6 (b-f) corresponds to the diagram of Fig.6A.

In Fig.6b schematically depicts a predefined matrix of choice WOOFER channel stored in the memory unit 130.

Dark area 46 corresponds to the singular values of the matrix elements, i.e., the spatial frequencies that are in the bass channel, and the light areas correspond to zero values of the matrix elements, i.e., those frequencies that are not in the LF channel.

In Fig. 6 (c-f) schematically depicts a matrix of selection of four RF channels. Used the same designations as in Fig. 6b.

It should be noted that the sum of all matrices of the choice of frequency channels (Fig.6, b-f) is the matrix whose elements are all equal to 1 (one), i.e., all information contained in the original image, at least one of the frequency channels.

The output of each of the matrix multipliers 120-124(Fig. 2) are formed of a matrix of the respective frequency channels in the frequency representation. To convert these matrices in the coordinate representation used inverse Fourier processors 140- 144.

Direct 11 and 14 reverse0- 144Fourier-processors can be made on the basis of the algorithm is troacoustics AU-17, 2, 77-86).

Examples of images generated at the outputs of the inverse Fourier processors 140- 144in the image processing Fig. 7a, shown in Fig. 7(b-f). The image WOOFER channel 7b output Fourier processor 140, which is the output 7 of the block division frequency 2, is fed to the input 9 of the block forming the output image, and the image of the four RF channels 7 (c-f) with the outputs of the Fourier processors 141- 144which other exits separation unit frequency 2, proceed to the appropriate inputs of the computing unit 3 correlations and to the inputs 101- 104the forming unit of the image output 5.

Consider the processing of this information on one frequency channel, since this processing is identical in all frequency channels.

The memory unit 15 (Fig. 3) is used to store the processed image frequency channel. To compute the non-normalized correlation values from the memory 15 sequentially selects the image element 21 and its environment 22. The latter is supplied to the multiplying input of the adder 17, which implements the following function:

< / BR>
where N is the number of image elements in the environment 22, preferably N=8, Vi- p is I from the environment 22. The weighted sum of neighboring picture elements is multiplied by the selected image element multiplier 18, the output of which receive the scaled correlation value for the selected image element. To compare the obtained correlation values (the output of block 18) with the threshold value (the output of block 20), the correlation is divided by the threshold value of the divider 19 and the result is compared to the unit in the processing unit weights 4. These operations are repeated for all picture elements of the processed frequency channel.

The image processed frequency channel is also fed to the input of block definition noise 20. Received in block 20 the result is used as a threshold to normalize the correlation divider 19. As a result, the output of the computing unit correlations 3 is a matrix containing correlation values for all image elements of a given frequency channel, and these values are normalized threshold value for a given frequency channel.

Obtained from block 3 of the correlation value by four channels arrives at the inputs of blocks rounding 231- 234(Fig. 4), which is the input block 4 formirovanie image.

Rounded data from outputs of blocks 231- 234arrive at the inputs of the block address generation 24, which converts the four four - or patibility words in one 16 or 20 bit. The generated address is supplied to the inputs of four functional converters 251- 254. Each of the functional converters 251- 254represents a block of memory in which are stored the values of the weighting factor for each combination of the four correlation values in the four frequency channels, and each combination corresponds to a single address value generated by block 24.

An example of a functional dependence of the weighting factor Wiany of the frequency channels from the correlation values Ciin this frequency channel and the correlation values in the other three frequency channels shown in Fig. 8, the threshold for that frequency channel. The weighting factor Widepends on the correlation values Ciand maximum values of L correlation in the other three frequency channels. This dependence is shown by the family of curves:

curve A if Ci0.7 L

curve B when Ci= 0.5 L

curve C with Ci1- 254accumulated in the memory blocks of weights 261- 264. The block address generation adders 27 and 281- 284serve to smooth these weighting coefficients in each of the frequency channels. To do this from memory block, for example 261sequentially selects the image element and its environment, which are received at the inputs of adder 281and folded. Their sum is supplied to the output unit 281one of the outputs of the shaping unit weights 4.

From the output of block 4 weighting factors are received at inputs 81-84(Fig. 5) multipliers 291-294which are the inputs of the processing unit of the image output 5, the second inputs 101- 104multipliers 291-294do the values of picture elements of four frequency channels from the outputs of the block division frequency 2. Received pieces of image elements frequency channels and their corresponding weighting factors are received at the inputs of the adder 30. In addition, the input 9 of the same adder receives the corresponding picture element of the low-frequency channel. The adder 30 produces the elements of the output image, which Acevedo an example implementation of the method as applied to the processing of two-dimensional scalar images. Similarly, processing of three-dimensional images. In the elements of the device that implements the described method increases the number of frequency channels used three-dimensional Fourier transform processor and to calculate correlations using the sum of 8 and 26 of the nearest neighboring elements.

The inventive method can also be used for processing vector images, particularly color, where the three components of the vector correspond, for example, the intensity of the three basic colors for this image element. For this scalar operations such as Fourier transformation and summation are replaced by the vector, and to calculate correlations applies the scalar product of the given picture element in the vector is the weighted sum of its neighboring elements, while the vector adder contains one normal (scalar) the adder for each component of the vector.

The method of image processing according to the invention has the following advantages.

First, the correlation between the selected image element and its neighbors can detect weak patterns on the background noise, which provides a high filing/P> The correlation between the selected element and its neighbors allows to apply the described method to a wide range of images, including color, black and white and three-dimensional images.

Secondly, the definition of thresholds by analyzing the distribution of the image elements allows to provide high-quality image processing with significantly different levels of noise without tuning parameters.

Thirdly, due to the use of the LF channel provides faithful reproduction of large parts of the image.

1. The method of image processing based on the initial image in the form of a matrix of picture elements, dividing the original image into n frequency channels, each of which contains a matrix of the same dimension as the original image, the edge detection image and generating the output image of the preceding n frequency channels based on the selected boundaries, characterized in that the separation of the image into frequency channels emit low-frequency channel and n - 1 high frequency channels, to emphasize the boundaries in each of the n - 1 high frequency channels for each picture element p is to be placed, compares the correlation value with the correlation values for the respective position of the image elements in the other frequency channels and the threshold for this channel, by comparing the results form for each of the picture elements of each of the n - 1 high frequency channels weighting factor, and generating the output image is produced by summing each of the picture elements of the low-frequency channel with all the relevant position of the image elements of the n - 1 high frequency channels, multiplied by the corresponding weighting factors.

2. The method according to p. 1, characterized in that a weighting factor for each image element of each of the n - 1 high frequency channels is determined by comparison of the respective correlation values with a threshold value.

3. The method according to p. 2, wherein the weighting factor associated with the correlation value and the threshold value as follows: weighting factor is minimal, if the correlation value is significantly less than the threshold value; weighting factor monotonically increases from the minimum to the maximum value, if the correlation value of sravnimogo values.

4. The method according to p. 2, wherein the weighting factor associated with the correlation value and the threshold value as follows: weighting factor is minimal, if the correlation value is significantly less than the threshold value; weighting factor monotonically increases from the minimum to the maximum value, if the correlation value increases to a second threshold value equal to the product of the threshold value by a predetermined coefficient; a weighting factor monotonically decreases from a maximum value to limit value, if the correlation value is greater than the second threshold value.

5. The method according to p. 1, characterized in that m of the n - 1 high frequency channels, where 2 m n - 1, differ from each other only in the direction of primary transmission.

6. The method according to p. 5, characterized in that a weighting factor for each image element of each of the m high-frequency channels is determined by comparing the corresponding correlation values with a threshold value and the values of the correlations for the corresponding position of the image elements in the remaining m - 1 high frequency channels.

7. The method according to p. 1, characterized in that the element and the tion.

8. The method according to p. 7, wherein the correlation value for each picture element is calculated by multiplying the given picture element on a weighted sum of neighboring elements.

9. The method according to p. 8, characterized in that m of the n - 1 high frequency channels, where 2 m n - 1, differ from each other only in the direction of preferred bandwidth, and when calculating the weighted sum of neighboring elements using anisotropic weight, and the direction of anisotropy corresponds to the direction of preferential transmission of the processed frequency channel.

10. The method according to p. 7, characterized in that the threshold value for each of the n - 1 high frequency channels is determined by analyzing the distribution of values of image elements of this frequency channel.

11. The method according to p. 7, characterized in that the threshold values for all frequency channels is determined by analyzing the distribution of element values of the original image.

12. The method according to p. 1, characterized in that the picture element is a vector whose components describe, for example, the brightness of the three primary colors (red, green, blue).

13. The method according to p. 12 different is this picture element on a weighted sum of neighboring elements.

14. The method according to p. 13, characterized in that m of the n - 1 high frequency channels, where 2 m n - 1, differ from each other only in the direction of preferred bandwidth, and when calculating the weighted sum of neighboring elements using anisotropic weight, and the direction of anisotropy corresponds to the direction of preferential transmission of the processed frequency channel.

15. The method according to p. 12, characterized in that the threshold value for each of the n - 1 high frequency channels determined by analysis of the distribution of absolute values of vectors representing the picture elements of this frequency channel.

16. The method according to p. 12, characterized in that the threshold values for all frequency channels determined by analysis of the distribution of absolute values of vectors representing the elements of the source image.

17. The method according to p. 1, characterized in that before the calculation of the weighting coefficients in at least one of the n - 1 high frequency channels to produce a smoothing of the correlation values for several neighboring image elements.

18. The method according to p. 17, characterized in that before averaging the values of the correlations produce nonlinear Preobrazhenie, and values significantly large threshold values decrease.

19. The method according to p. 1, characterized in that when forming the weighting factor in at least one of the n - 1 high frequency channels to produce a smoothing of this weighting factor for several neighboring image elements.

20. The method according to p. 1, characterized in that the input image is a p-dimensional matrix of picture elements, where p 3.

21. The method according to p. 1, characterized in that for forming the weighting factor to at least one of the n - 1 high frequency channels use different thresholds for different parts of the image.

22. The method according to p. 21, characterized in that the picture element is a single number, and the threshold values for various parts of the image and different frequency channels determined by analysis of the distribution of element values of the respective parts of the image corresponding frequency channel.

23. The method according to p. 21, characterized in that the picture element is a vector, and thresholds for different parts of the image and different frequency channels determined by analysis of raspredelenie frequency channel.

 

Same patents:

FIELD: computer science.

SUBSTANCE: device has image sensor, block for separating frame and string pulses, generator, analog-digital converter, and also has digital signal processor and block for controlling executive implements of robot. Digital signal processor performs processing of image, going as follows: increase of brightness and contrast of image, transferring semi-tone image to two-gradation one, singling out contours, objects, determining position of objects. On basis of received data control block for robot implements drives trajectory of robot displacement, selects necessary command and forms controlling signals.

EFFECT: higher efficiency.

1 dwg

FIELD: technology for biometric control and authentication.

SUBSTANCE: on basis of registered biometric data only those characteristic data are received, which are unambiguously reproducible. Then on basis of these unambiguously reproducible characteristic data additional set of access key parameters is generated, which represents personal identification code, or on basis of which such an identification code can be generated.

EFFECT: when implemented for recognition and reading of biometric data, fingerprints in particular, higher reliability and resistance to interference during identification of persons.

4 cl, 1 dwg

FIELD: control-measuring criminalistics equipment, possible use for determining remoteness of sweat-fat hand marks.

SUBSTANCE: method for determining remoteness of sweat-fat hand mark includes detection of mark, for example by means of dactyloscopic powder, photometry of image of papillary lines in all mark points, digitization of image of mark together with background surrounding it, construction of brightness histogram relatively to mark and relatively to background, determining of average quadratic brightness deviation relatively to mark and relatively to background, production of their relation and comparison to calibration curve produced using a test object.

EFFECT: increased precision when determining remoteness of mark.

7 dwg

FIELD: systems for automatic video surveillance of an object.

SUBSTANCE: system for automatic detection and tracking of individuals on basis of images and biometric identity recognition based on target list, realizes following operations: on basis of three-dimensional data about scene and two-dimensional data, characterizing optical flow, detection of objects-noises of scene is performed, static background objects are selected, and regular dynamic object-noises; on basis of comparison of two-dimensional and two-dimensional data about the scene in current frame with reference data on previous frames and a map of object-noises changes are determined on a scene, in three-dimensional zones of interest, preliminary check of presence of human-like objects is performed, zones of interest are determined more precisely and their changes are tracked: a contour of separate elements of human body is singled out, zones of interest are divided onto a set of sub-zones of interest for elements, detection of three-dimensional head of individual is performed and it is tracked in each zone of interest; face of individual is tracked in each zone of interest; images of detected face are normalized in terms of dimensions, angles and brightness; recognition is performed.

EFFECT: objectivity and stability of system operation.

1 dwg

FIELD: machine vision systems.

SUBSTANCE: gradient method of object contour extraction at a halftone raster image matrix is proposed. The substance of the method is as follows: for all raster image pixels, the pixel brightness gradient norm or norm square is calculated using the selected method; then all elements, for which the gradient norm or the norm square exceeds the threshold value, are highlighted black on white on a new black-and-white monochrome matrix; coherent configurations of black elements are considered as object contours on the monochrome matrix; the factor is determined for the selected gradient calculation method, then the gradient norm square threshold value is calculated as a product of this factor times the sum of mean adjacent pixel brightness change modulus value squares, by rows and columns, for pixels, the values for which exceed the grand mean non-zero change levels, by rows and columns accordingly; among coherent configurations of black elements on the monochrome matrix, configurations with less than 5-7 elements are rejected immediately; for the remaining configurations, the mean degree of adjacency is calculated as a quotient of the sum for all configuration elements adjacent to that element divided by the sum of elements in the configuration; configurations with a mean degree of adjacency less than 3 are rejected, and the remaining configurations are accepted as the required object contours.

EFFECT: increase in object contour extraction quality in raster images.

3 dwg

FIELD: physics.

SUBSTANCE: method of grey image conversion colour is offered that implies that the original grey image presented elements array is divided on n frequency channels. Low-frequency and n-1 high-frequency channels are isolated and correlation of elements within high-frequency channels is calculated that is used a basis to calculate weight factors of elements within high-frequency channels. Each frequency channel is given with colour, and output image is created by element-specific sums of colour of each frequency channel assuming element weight factors within high-frequency channels. At synthesis of output image, at least, in one n-1 high-frequency channel part of weight factors are nulled and/or smoothed.

EFFECT: expansion of image processing possibilities.

4 cl

FIELD: physics, image processing.

SUBSTANCE: invention concerns digital image processing and can be applied in devices for automatic or personal identification by fingerprints. The method includes intensity gradient assessment calculation for each image pixel, including calculation of average intensity gradient magnitude for each pixel in arbitrary shaped local image areas with their size defined by parameters set in advance, calculation of average gradient magnitude for processed image, division of all pixels into two groups (background and image), object and background areas tagging and defining their size, forming of continuous areas by uniting tagged areas and defining their sizes, defining size limit values for continuous object areas, defining size limit values for continuous background areas, transition of object and background areas with sizes less than product of obtained limit values and given weighting factors to background and object areas respectively.

EFFECT: improved quality of object areas selection.

4 dwg

FIELD: physics, computation equipment.

SUBSTANCE: the invention claims method of image blur compensation involving: calculation of difference between measured image pixel brightness and brightness assessment obtained earlier on the basis of previous frame sequence; movement detection by comparison of obtained difference to threshold value; defining of movement direction for each pixel; combination of adjoining pixels with the same movement direction in a single object; outlining contours of moving objects by adding their initial B(k) and gradient ▿(B(k)) of images; and generation of output image where k1, k2 are weight factors. Device of image blur compensation includes: image sensor, controller, mode movement detection module, object detection module, correction module, first RAM device, second RAM device, third RAM device, counter, first comparator, second comparator, first multiplexor, second multiplexor, third multiplexor, fourth multiplexor, fifth multiplexor, sixth multiplexor, seventh multiplexor, first demultiplexor, second demultiplexor.

EFFECT: blur compensation for moving object image in real-time mode.

2 cl, 11 dwg

FIELD: information technology.

SUBSTANCE: present invention relates to generation and use of files, containing mixed raster. The method of forming a mixed raster content (MRC) image involves stages for receiving commands for generation of an image file form an application program in form of a MRC image; generation of said MRC image by using a printer driver; generation of a selective layer; generation of a combined main/background layer; generation of lists of points for each region of the image, including the main layer, background layer and selective layer, where the lists include one or several of the following points: bounding region, type of graphic output and used colour; processing the said lists, where if the region of the image is text or line drawings, the region of the image is added to the selective layer, otherwise the region of the image is added to the main/background layer; and generation of a background layer and a main layer, each of which includes the corresponding point in the list of the background layer and list of the main layer, which does not intersect with the point in the list of the selective layer.

EFFECT: cutting on coding, decoding and composition time.

12 cl, 20 dwg

FIELD: protection equipment.

SUBSTANCE: invention relates to biometric control and identification. For this purpose, fingerprints of a person are recorded using a sensor and initial data are formed. Characteristic data are obtained, which are chosen taking into account statistical confidence of identifying a person and a set of access key parametres are generated. At least three fingerprints are recorded, which are then used to generate at least two sets of extended access key parametres. The system for verifying identity of a person from fingerprints has sensors, a device for forming initial data, a device for obtaining characteristic data, device for generating a set of access key parametres, a device for comparing set of access key parametres with standard data, and a data output device. The system also has a device for generating a set of extended access key parametres. The data output device identifies a person as having access rights, if there is a complete match in the comparing device of at least, one set of extended access key parametres with the set of standard data.

EFFECT: increased selectivity of access key, increased probability of unambiguous generation of access key parametres and increased reliability of the system.

6 cl, 1 dwg

Up!