Method for selecting an image fragment for television coordinators

FIELD: television engineering, in particular, method for selection of objects on complex underlying background, possible use for systems for automatic detection of coordinates of objects in television automatics.

SUBSTANCE: in accordance to the method, histogram processing threshold is optimized with the target of reaching minimal radius of compactness of binary image. For that purpose, a set of weighted radiuses of binary elements is computed and resulting compactness radius is compared to standard, which represents a radius of ideally compact figure, such as a circle with area equal to area of resulting binary image. Then, range of one of histograms is reduced or increased and compactness radius calculations are repeated until its value maximally approaches the standard value. Object coordinates shifting value is determined by scanning the area of object/background binary image with sliding strobe of object and by detection of position with the least compactness radius.

EFFECT: high selectivity of object image.

1 cl, 5 dwg

The invention relates to television technology, in particular to a method of selecting objects on a complex underlying background, and is intended for automatic positioning of objects in various areas of television automation.

There are various methods of selection of images of objects or fragments of images in the complex underlying background. These methods involve stepwise (sequential) selection in individual parameters of the field of object-background. Among the most common are selection methods using two-dimensional gates that restrict the image size to avoid redundancy of data to be processed. As a subsequent stage of breeding successfully use the method of binarization of the image. On the basis of the brightness or contrast form a new image, the elements of which assigns a value of "1"if this item refers to the breeding facility, and "0" otherwise. Gates typically have a rectangular shape with smaller sizes for selection of the object and with a large selection object/background. Known trick is the optimal framing of the image of the object, i.e. to reduce the sizes of the gate in accordance with the image size of the object so that the inside of the gate to minimize the number is in the picture elements, belonging to the background. This decision was made in the device "Television coordinate discriminator", USSR author's certificate No. 1700766, CL N No. 7/18. Here, in order to adapt the dimensions of the gate to the size of the object device generates a series of gates with discrete measurements. Along the perimeter of each of the gate frame is formed with a thickness in a single item. Condition adaptation is filling gate binary signal, but within the peripheral gate (frame) binary elements must not contain. Processing is performed in all the gates simultaneously. Automatically selects the size of the gate, in which the above condition is satisfied.

The disadvantage of this discriminator is that the binary signals for a wide class of observed subjects not so compact, in order to meet the set requirements. Such is the case with the more likely the lower the contrast between object and background. The elements of a binary image can be scattered in the gates and the condition may not be fulfilled. In addition, the square shape of the gates for most cases not optimal and the formation of many rectangles with different aspect ratios considerably reduces the effectiveness of this principle both in hardware cost and execution time.

WPI is local correlation methods selection. The gate object record standard, and within the gate background of the search object, displaced as a result of its movement in relation to the background or bias field of the object-background due to the shift of the image relative to the television raster. The search carried out by the criterion of maximum similarity between the reference and current image. Correlation method uses a multi-level image processing. Along with its obvious advantages correlation method has the disadvantage that the brightness and contrast methods are deprived. Therefore, correlation methods complement the present description of the selection methods.

Effective method of selecting the brightness of the object in the background is a method that uses the luminance histogram. This method is described in the article "image Segmentation: thresholding methods" in the journal "Foreign Radioelectronics No. 10" for 1987, P6-24.

In this method, the image segmentation involves the analysis of the histogram, when fashion brightness of the background and object distinguishable. The histogram depicts the curve, which in the particular case of two modes of brightness: object and background, between which there is a depression, conditionally designating the boundary of the object background. Intensity modes of brightness depends on the ratio of the area occupied by the background and the object according to the respectively, A priori, this characteristic of the image is not known, so the threshold is set at the trough. In this limit, set the threshold to form a binary image by assigning the value "1" brightness elements belonging to the object, and "0" to the others.

As a General comment, one can point to the ambiguity of thresholding in mismodeling depression histograms, because in this area with almost the same probability included background elements and lost the parts of the object.

Another example, when the thresholding processing ambiguous - this is a case with a tri-modal histogram is the case when the background contains contrasting changes in brightness, and the brightness of the object is different from them. Then the problem of selection will need to install at least two threshold that is twice the ambiguity threshold processing.

Closest to the proposed invention similar to that adopted for the prototype, the technical essence is a method of image segmentation, described in the article "Television tracking system with Bayesian discriminator goals", published in the journal "Foreign Radioelectronics No. 10" for 1987, P.81...91. In this way the selection of the image of the object produced by the method of segmentation of individual on-site gate, providing for the filling of the gate is the image of the object This allows you to completely isolate the subject from the background. Use methods multithreshold processing and formation of a binary image. Maintaining the constancy of the ratio of the size of the object/background in the prototype provide the boundaries of the object in the binary image by selecting them from four sides of a single witness gates and calculating in each cycle pairwise divergence between the position of the gates separately for each axis. If necessary, adjust the size of the gate so that the object boundaries coincide with the boundaries of the gate.

However, it is clear that the image of the object contains non-uniformity of the brightness, including matching with elements of the background brightness. Therefore, the binary image may not have distinct boundaries, especially at low contrast object/background. In addition, during the automatic tracking of the object strobe occurs, the offset between the position of the gate in the previous TV shot (TV box) and a new position in the current. This offset is greater, the greater the angular velocity (and in some systems, angular acceleration) of the optical axis of the television camera.

In the prototype injected inertial update histograms, resulting in a time histogram is less responsive to the appearance of the object area of the background elements. In other words, reduce the risk of appearing in the binary image elements with the values "0" in the object area and "1" in the background area. But still remains uncertain quantitative measure of inertia updates. A small degree of inertia may be less effective, on the contrary, deep inertia will be harmful to rapid changes in theobjectives situation.

To fill a single "zero" elements "holes" in the area of the object and the exception of a single "unit" excessive "false" elements in the background region in the prototype used a special filter element aperture. Naturally (this is written by the authors), this filter reduces the edges of the size of the binary object image and for his recovery requires one filter to another filter function. For processing mentioned a single item, you need a filter with aperture 3×3 items. The same size filter you will need to restore the boundary elements. It should be noted that the appearance of a single false "1" and "0" the event is not typical for a high-quality image. The false nature of the entities included in the correlation images as backgrounds and objects. In this case, the filters that eliminate false two-dimensional education, with a long form would require more hardware, but it's a different class of complex image processing.

We propose a method of selecting a portion of the image is La television coordinators, involves pre-processing of the video signal by digitizing and histograms of the elements of the image brightness and the introduction of the circuits of their inertial upgrades within the area of the gates, through which selectrow image of the object and its surrounding background separately, the thresholding (threshold) selection, which form the binary image, and determine the increments of the coordinates of breeding fragment relative to the previous values of its coordinates, for example, by the criterion of the "center of gravity", then calculate the radius of the compact binary image enclosed within the gate fragment, using the weighted sum of the radii of binary elements, compare it with the standard, which represents the radius of the ideal compact shape which is a circle with an area equal to the square of the obtained binary image, then gradually reduce or increase the scope of one of the histograms and repeat the calculation of the radius of the compact up until its maximum value is not close to the reference value, and then determine the offset value (incremental) coordinates of the object by scanning the area of the object/background binary image sliding gate object and finding the position with the smallest radius of compactness. In ka is este criterion of maximum proximity values of the radius of the compact to the magnitude of the reference radius accept the result, which is repeated several times, or when the radius of the compact from frame to frame changes less than the given value. The constancy of the relationship of the dimensions of the gate object to the size of the object, characterized by the value of its compactness, save by scaling the current image, and scaling the current image is produced on each axis separately by defining a vector of compactness of the object in different directions of the coordinates. Inertial updating histogram of the elements of the brightness of the object image is carried out in dependence on a value of frame-to-frame displacement of the image coordinates of the object.

The technical result is to obtain high selectivity image object.

The technical result is achieved due to the fact that in the process of selection of the object image by the method of histogram processing, thresholding selection for fashion brightness and forming a binary image analyze the compactness of a binary image of the object and upon the results smoothly change the value of threshold selection to obtain in the area of the object as a compact image of the object, then adjust (by adjustment) size selectivity of the gates to the size of a compact binary image of the object, for example the EP, by scaling and centering of the image of the object relative selectivity of the gate.

Figure 1 shows the block diagram of the algorithm that implements the proposed method of selection of an image for television coordinators.

Figure 2 illustrations of the phases of image processing: ensure the compactness of a binary image, to adapt its size to the dimensions of the gate and use the most compact binary image to determine the increments of the coordinates of the object.

In the proposed method the criterion of adjustment are not the object boundary, because clear boundaries do not exist, and the compactness of a binary image. When this adjustment of the size of the gate is carried out indirectly, i.e. not change the size of the gate, and the scale of the original image, leaving constant the size of the gates that provides a constant run-time processing when implementing the method. To scale separately determine the compactness of the axes of coordinates, which allows to optimally frame the object gate. In addition, inertial update histograms produced only on the histogram object. This is necessary because the witness gate to a greater or lesser degree stores information about the object within it, while the underlying background can sonatica, for example, due to the movement of the object on it.

As a measure of the inertia of the pack in the proposed method uses the values of the interframe displacement of the object image. For example, the product of two quantities of offsets on coordinate axes, expressed in the same units of measurement as the gate (in the elements of the square gate), can be successfully used as a measure of the inertial update, showing how the object came from outside the gate, and liberated the area occupied background elements. In these short periods of time increase the time inertial pack, and when the gate once again is filled with the image of the object, speed up the upgrade process.

As shown by the results of mathematical modeling, this measure objectively regulates the inertial time update of the histogram.

Elimination of heterogeneity binary image for its idealization by suppressing some and recovery of other elements of a binary image is almost impossible. In the end, by definition, the image of the object can exist elements of equal brightness with the background elements. In this case, when determining the coordinate position of the object is important to have not only (and not so) complete the shape (silhouette) of the object. More important condition selection using histograms is the most in what you select elements of brightness, belonging exclusively to the object. This form of the object is more stable. In the gate object is the minimum number of background elements. This contributes to the proper determination of the displacement coordinates of the object, for example, in the "center of gravity" of the lamina, therefore, a more accurate determination of its coordinates.

In the proposed method uses multiporous processing histograms. It consists in the following. Each value of p the intensity of the brightness in the histogram Ho_pthe object is compared with the value of the same intensity in the histogram Hf_pbackground. In this form the vector Hys_p:

- if But_p>Hf_pthen position R vector Hys_passign the value "1";

- otherwise "0".

Then the vector Hys_pform a binary image as follows. For each of the intensities from 0 to p_maxby analyzing the original image In_{m, n}consider the condition In_{m, n}=Hys_p. If equality occurs, then the binary image Bb_{m, n}assign "1", otherwise "0".

When forming the histograms should be taken into account that the area of the gates of the background and object are often not the same. This means that it should be according to some criterion (for example, the ratio of areas) to normalize their intense the property.

On the one hand, the presence of elements of brightness relative to the background and the object, indicates that the histograms overlap. Then the estimated normalization would be ineffective without a priori data about the overlap of the histograms.

On the other hand, if we assume a decrease of the amplitude histogram of the object, there is a favorable trend towards a reduction in General for both overlapping histograms of brightness values. In this case reduces the number of elements in the binary image that belong only to the object, but at the same time and decreases the number of binary elements within the background region. This approach provides a more compact focus the elements of a binary image object area (even if, in the particular case, price reduce its own size).

To provide an objective reduction of the amplitude histogram of the object (or, equivalently, increasing the scale of the histogram of the background) in the proposed method uses a criterion of compactness image Bb. In this case, calculate the compactness and compares it with the target value chosen close to the ideal (reference) value. Change (adjust) the scope of the histogram is performed in stages, controlling changes compactness. The date of termination of the adjustment determined either by reaching the plant values, or if adjustment of subsequent phases does not significantly affect the calculated compactness.

In the prototype, the compactness is defined as

where S is the total area of the binary image;

L is the total perimeter of all the individual entities of a binary image.

This definition is empirical and true for deterministic images having a limited proportion of sizes. Accidental formations binary image definition of compactness may be ambiguous due to the large scatter between the values of S and L dispersed binary image that can negatively impact both in the framing of the object by the gate, and when determining its center.

In the proposed method, the compactness of a binary image is determined as follows:

where Rc is the average weighted radii drawn from the center coordinate of the gate to the center element of a binary image.

where m, n variables by coordinates;

Bb - element binary image;

Sb - the total area of the binary image

Ret - radius shape with ideal compactness of a circle with radius Ro, the area of which is equal to Sb. However, the radius of him to whom mactest differs from Ro.

The value of Ret in the General case can be calculated by the formula

where dS(R) is the space of elementary shapes summed (integrated) from a smaller number to the maximum So, for example square elementary figures presented in the form of rings with radius R and thickness dR, so

Integration of produce from zero radius rings up to a maximum radius Ro. Therefore, substituting (2) into (1), and then driving like that, we get

Integrating the numerator and denominator of (3), we obtain

The sequence of actions to implement the proposed method is in accordance with the block diagram of the algorithm is shown in figure 1.

In block "1" to produce a digitized video signal in a particular area of an image, the size of which is greater than the specified size of the gate background. The ratio of these dimensions is determined by the adopted multiplicity scaling in the implementation of adaptation of the size of the object to the size of the gate object. This ratio has its limits, because when you zoom in any direction, there occurs a redistribution of brightness due to the emergence or disappearance (compression) of the image elements. In the compact region shall be a binary image changes its shape and size.

In block "2" set initial scaling factors, and block 3 carries out the scaling factor, arriving at its inputs.

Block 4 distributes the gates. If the gate object in the block "5" form a histogram of the object But_potherwise the gate of the background in the block "6" form a histogram of the background Hf_p.

In the block "7" set ratio "a", which regulate the scale of the histogram But_p. The initial value of this parameter is not strictly determined by the ratios of the areas of the gates of the object and the background. In this example, conditionally accepted "a=1". In the block "8" produce the actual change of the amplitude histogram.

Block 9 is a comparator that determines which levels of brightness of the object predominate over the same background levels. Comparison build vector Hys_pwho in the block "10" assigned "1"if the condition is met, and in the block "11" assign "0" otherwise.

In block 12" compare image elements In_{m, n}with vector Hys_pand the results form the binary image Bb_{m, n}. If the brightness level of the image coincides with one of the single values of the vector Hys_pthen in the block "13" image Bb_{m, n}assign the value "1", otherwise, in block "1" image Bb _m,nassign "0".

In the block "15" calculates the compactness of a binary image in accordance with the formula

and in the block "16" compare her with some value δ>1, taking into account that random compact shape may not be perfectly compact. As long as the value of compactness exceeds the set value δin the block "17" check the change of compactness in the i-th frame with respect to the compact obtained in the (i-1)-th frame. If further changes compactness exceed some specified value Δthen in the block "18" correcting coefficient "and" decrement da<1 in order to reduce the scale of the histogram object. If the changes compactness does not exceed a given value of Δthen in blocks "19" and "20" carry out m times checking, repeating the correction coefficient "a", then move to the calculation unit 21 increments of coordinates ΔX, ΔY and X, y In the block "22" carry out the adaptation of the size of the object to the constant size of the gate object. In the "23" select the channel processing. If Y, then in the block "24" determine the compactness of Y, and in the block "25" compare her with some value δ'y. If the condition Comp Y more δ′y, then in the block "26" decrease the zoom factor My multiplier νy<1, otherwise in BC is ke "27" Comp Y is compared with the value δ ″y. If the condition Comp Y is less than δ″y, then in the block "28" increase the zoom factor My multiplier μy>1. If none of the conditions is not met, i.e. if the value of Comp Y remains within δ′>Comp Y>δ″y, then My remains unchanged. If the processing of the channel X, the actions of blocks "24"..."28" similarly transferred to the blocks 29"..."33". In this example, the scaling factors are generated by compact binary image for each axis separately. There may be other ways to determine Mx and My, which can be successfully applied in the framework of the proposed method.

Figure 2 is an illustration of a sequence of actions to achieve maximum compactness of a binary image, adapting the dimensions of the compact region to the size of the gate and estimate the offset of this field relative to the current position (or the center of the coordinate system for the background area).

In the initial moment of time, when the size of the object is not adapted to the dimensions of the gate, within the gate object get the brightness elements that belong to both object and background. In this case, the binary image takes the form shown figa, the compact is small and conditions satisfactory compactness are not met. Therefore, adjusting the coefficient "a".

N is figb after the correction of the amplitude histogram of the object decreased the number of binary formations, belonging to the background, even at the price reduction and "cutting" the area of the object.

On FIGU produced adaptation of the size of the binary object image to the dimensions of the gate by scaling the original image along two axes separately.

On Figg shows how to define a coordinate increment ΔX, ΔY by scanning the gate object just background space. This fixed the position coordinates of the gate object, which is the most compact part of the binary image.

The practical implementation of the proposed method is possible on the basis of modern technologies, in particular microprocessors and programmable logic integrated circuits (FPGA). The accuracy of the possibility of realization is confirmed by the results of mathematical modeling in MathCAD. The model consists of two branches. In the first branch simulates the original image, the second implements the algorithm shown in figure 1. Model image consists of a random field brightness with the given statistical characteristics (dispersion, the correlation interval and others)that simulates a moving background. The background contains abrupt changes in brightness, simulating abrupt transitions, for example, from a brightly lit area of the earth surface in shadow or the intersection of the horizon line. On the background of the attack of the woman, the image of the object with a random luminance relief, which are brighter or darker than the underlying background is also moving relative to the background in different directions diagonally. As a result, the object does motion relative to a moving background in different directions, crossing the contrasting differences of background.

Model of the proposed method is implemented completely except for the adaptation of the size of the object/background. This part of the algorithm was implemented separately. In the model investigated various characteristics of the inertial update the histogram object, methods of scaling and determining the center of the object along the compactness of its binary image.

The simulation results confirmed the viability of this method when using the proposed formulas for calculating the compactness for scanning by "sliding" the gate, determine the increments of coordinates ΔX, ΔY, use these values to control circuits inertial upgrade and use scaling to adapt the size of the object/background.

Probabilistic assessment of the results shows that the proposed method of selection provides almost two times higher reliability of selection of an object on the complex underlying background and in the same lower fluctuation error in comparison with the known methods of selection.

1. The method of selection of an image for television coordinators, consisting of pre-processing of the video signal by digitizing and histograms of the elements of the brightness of the image and inertial update within the area of the gates, selectivity image of the object and its surrounding background separately, the thresholding (threshold) selection, the formation of a binary image, the subsequent determination of the increments of the coordinates of the fragment relative to the previous coordinate values, for example according to the criterion of the "center of gravity" of an image, characterized in that calculates the radius of a compact binary image enclosed within the gate snippet using the weighted sum of the radii of binary elements, compare it with the standard representing the radius of the ideal compact shape in the form of a circle with area equal to the square of the obtained binary image, smoothly reduce or increase the scope of one of the histograms and repeat the calculation of the radius of the compact up until its maximum value is not close to the reference value, and then determine the offset coordinates of the object by scanning the area of the object/background binary image sliding gate object and finding the position with the smallest radius of whom is Aktoty.

2. The method according to claim 1, characterized in that as a criterion of maximum proximity values of the radius of the compact to the magnitude of the reference radius accept the result, which is repeated several times in a row or when the radius of the compact from frame to frame changes less than the given value.

3. The method according to claim 1, characterized in that the constancy of the relationship of the dimensions of the gate object to the size of the object, characterized by the value of its compactness, save by scaling the current image.

4. The method according to claim 3, characterized in that the scaling of the current image is produced on each axis separately by defining a vector of compactness of the object in different directions of the coordinates.

5. The method according to claim 1, characterized in that the inertial updating histogram of the elements of the brightness of the object image is carried out in dependence on a value of frame-to-frame displacement of the image coordinates of the object.