RU2552195C1 - Organisation of image spatial filtration core and device to this end - Google Patents

Abstract

FIELD: physics.

SUBSTANCE: in compliance with this method, binary image of initial image fragments is generated by linear convolution with the help of spatial filter, its core including central and peripheral parts. Note here that linear convolution is conducted separately with spatial filter central core and linear parts. Major components used as binary images of initial image fragments are constructed wherein spatial filter central part features an exponential shape. Note also that peripheral part has negative factors and is constructed with the help of exponent proceeding from the sum of said central and peripheral parts being zero.

EFFECT: higher quality.

2 cl, 5 dwg

Description

The invention relates to the field of processing images formed, in particular, by rulers and matrix photodetectors, and can be used to filter images in automatic recognition systems to normalize and center image histograms.

A known method of filtering images [RU 2407056 C2, G06T 5/20, G06K 9/40, 07/02/2007], in accordance with which the first passband (PP) is determined based on the data of the original image, the filter coefficient matrix (CF) is calculated to implement the frequency characteristics corresponding to the restriction on the frequency band (IF), using the first PP, the first filtered image data is generated by filtering the original image data using the matrix of the first KF, objective quality estimation values are obtained and the image data of the first filtered image and the calculation of the distribution coefficient (RR) used to determine the optimal PP, based on the value of the assessment of objective image quality, determine the optimal PP corresponding to the calculated CR by referring to the table in which the corresponding relationship between the RC and the optimal PP is determined , calculate the matrix of optimal KF for the implementation of frequency characteristics corresponding to the limitation on the IF, using the optimal PP, and generate data optim lnly filtered image by filtering the source image data using the matrix of optimal KF.

The disadvantage of this method is the relatively narrow scope, due to the fact that the method has a limited ability to be used in automated recognition systems, since it does not allow to form a binary image of the original image, which is most preferred for real-time image recognition.

Closest to the technical nature of the proposed is the method [RU 2494568, C2, H04N 7/32, G06K 9/40, 09/27/2013], which contains a determination step, which determines in accordance with the number of taps of the filter used for filtering, the number pixels across the width of the strip located outside the macroblock including the reference block, which is a block of the decoded reference frame, and which is in contact with the specified reference block, the step of obtaining at which the specified reference block and the strip corresponding to The pixels found in the determination step if the reference block, which is a block of the reference frame corresponding to the block included in the image to be filtered, is in contact with the boundary of the macroblock including the reference block, and the filtering step is performed filtering the image of the reference block and the strip obtained in the acquisition step, while in the filtering step, low-pass filtering is performed with respect to the difference image between the plurality of images, filtering the high frequencies with respect to the image obtained by the low-pass filtering performed by the first filtering means, and adding the image obtained by the low-pass filtering performed by the first filtering means and the image obtained by the high-pass filtering performed by the second filtering means, to any one of a plurality of images to generate a predicted image in units of macroblocks.

The disadvantage of this method is the relatively narrow scope, due to the fact that the method has a limited ability to be used in automated recognition systems, since it does not allow to form a binary image of the original image, which is most preferred for real-time image recognition.

The problem to which the invention is directed with respect to the method is to expand the scope by providing the ability to normalize and center the histograms of the source images, which, in turn, provides high-quality formation of a binary image of the source image.

The required technical result consists in expanding the scope of the method by introducing an additional arsenal of technical means (method operations), allowing to form a binary image of the original image, which is the most suitable for real-time image recognition.

The problem is solved, and the required technical result is achieved by the fact that in the method based on the sequential formation of fragments of the original image in the form of discrete matrixes of video signals, according to the proposed method, a binary image of fragments of the original image is formed by their initial linear convolution with a spatial filter, the core of which contains a central and peripheral parts, while the initial linear convolution is carried out separately with the central and linear parts of the spatial core filter, the results of which form the modules of their difference, used as binary images of fragments of the original image, the central part of the spatial filter has an exponential form, and the peripheral part has negative coefficients and is formed using an exponent based on the condition that the sum of the central and peripheral parts is equal to zero.

Devices for filtering images are also known.

One of such devices [RU 2354071, C2, H04N 5/14, 07.24.2009] contains a series-connected analog video signal sensor, an analog-to-digital converter, a unit in which, using amplifiers and adders, the brightness value of the filtered image pixel is calculated, a unit that determines the moment of the beginning and end of the processing of the frame, as well as the control unit for the amplification factors of the amplifiers.

The disadvantage of this device is the relatively narrow functionality, due to the fact that the device has a limited ability to be used in automated recognition systems, since it does not allow to form a binary image of the original image, which is the most suitable for high-speed image recognition.

The closest in technical essence to the proposed one is an image processing device [RU 2494568, C2, H04N 7/32, G06K 9/40, 09/27/2013], containing determination means for determining, in accordance with the number of taps of the filter used for the filtering process, the number pixels across the width of the strip located outside the macroblock including the reference block, which is a block of the decoded reference frame, and in contact with the specified reference block, means for obtaining from the reference frame the specified reference block ka and strip corresponding to the number of pixels determined by the determination means, if the reference block, which is a block of the reference frame, corresponding to the block included in the image to be filtered, is in contact with the boundary of the specified macroblock including the reference block, and filtering means for filtering the image of the reference block and the band obtained by means of obtaining, while the filtering means include first filtering means for filtering low frequencies in relation and a differential image between the plurality of images, the second filtering means for filtering the high pass in relation to the image obtained as a result of the low pass filtering performed by the first filtering means, and an adder for adding the image resulting from the low pass filtering performed by the first filtering means, and the image resulting from the high-pass filtering performed by the second filtering means to any of a plurality of images for generation is predicted my image in units of macroblocks.

The disadvantage of this device is the relatively narrow functionality, due to the fact that the device has a limited ability to be used in automated recognition systems, since it does not allow to form a binary image of the original image, which is the most suitable for high-speed image recognition.

The problem to which the invention is directed relative to the method is to expand the functionality of the device, aimed at providing the ability to normalize and center the histograms of the source images, which, in turn, ensures high-quality formation of a binary image of the source image.

The required technical result consists in expanding the functionality by introducing an additional arsenal of technical means, which enable normalization and centering of the histograms of the source images, which, in turn, provides high-quality formation of a binary image of the source image, which is most suitable for high-speed image recognition.

The problem is solved, and the required technical result is achieved by the fact that in the device containing the block of the reference frame in the form of a discrete matrix of video signals, as well as filtering means, according to the proposed device, a group of amplifiers is introduced, the input of each of which is connected to the corresponding output of the block of the reference frame, and filtering means contain central and peripheral parts, each of which is made in the form of an adder with weighted inputs, and an algebraic adder, while the inputs of the adders with weighted the inputs are connected to the outputs of the amplifiers, and the outputs are connected to the inputs of an algebraic adder, the output signal module of which is the output of the device.

The drawing shows:

figure 1 - diagram of a device for implementing the method of organizing the core of spatial filtering;

figure 2 - the core of the spatial filter;

figure 3 - the Central part of the core spatial filter;

figure 4 - the peripheral part of the spatial filter;

figure 5 - source image in various spectral ranges and presented below them, their histograms and images with histograms after filtering them.

A device for implementing the method of organizing the spatial filtering core (Fig. 1) contains a block 1 of a reference frame in the form of a discrete matrix of video signals (a special case is presented in the form of 9 photosensitive elements (photodiodes).

The device for implementing the method of organizing the spatial filtering core also contains a group of 2 amplifiers, the input of each of which is connected to the corresponding output of the block 1 of the reference frame, and filtering means containing a central part in the form of an adder 3 with weighted inputs 4 (resistors when executed in analog form ) and the peripheral part in the form of an adder 5 with weighted inputs 6 (resistors when executed in analog form, as well as an algebraic adder 7, the output signal module of which is the output devices, while the inputs of adders 3,5 with weighted inputs are connected to the outputs of the amplifiers 2, and the outputs are connected to the inputs of the algebraic adder 7.

A device is operating to implement a method for organizing a spatial filtering core as follows.

Images formed by modern matrix photodetectors and rulers are oriented to storage and their subsequent visualization to the human operator. At the same time, considerable efforts have recently been made to use these images in the interest of solving the problems of automatic detection and classification of objects of given classes.

Typically, such images are pre-processed, in particular binarization or fragmentation. To obtain a satisfactory binary image, the histogram of the original image must be centered. This transformation is reduced to a linear convolution of a special kind of kernel with the original image.

As studies have shown, the histogram of thus transformed images tends to a one- or two-mode look. Moreover, the optimal binarization threshold from the point of view of a significant difference in the corresponding average values can be selected from a whole interval. In this case, a pronounced robust character is achieved, and thereby an acceptable quality of binarization is ensured in a wide range of changing phono-noise conditions. For definiteness, images with a centered one- or two-mode histogram will be considered normalized.

Images obtained by matrix photodetectors or arrays in different spectral ranges are, as a rule, not centered and do not have a single or two-mode nature, which indicates non-optimal exposure time of the image on the photodetector and / or insufficient dynamic range of the photodetector. To normalize such images, a spatial filtering core of a special kind is proposed, a graphical image of this core is presented in figure 2. This kind of core is due to the requirements for smoothness and continuity of the integral properties of the core.

The core of the spatial filter consists of two components of the P-center and M-periphery, which are included among themselves in differential mode. The core is composed of two parts to provide spatial differentiation of the image in the local basis, which allows you to remove the constant component of the image. A necessary requirement for such nuclei is the vanishing of the sum of the P-center and M-periphery. Thus, the core belongs to the class of wavelet functions in terms of the interaction of its two parts of the P-center and M-periphery. The components of the spatial filter core are shown in FIGS. 3 and 4.

A linear convolution of the spatial filter core Q = [q (i, j)] of size m × n with an image S = [s (i, j)] of size k × d allows you to generate an image U = [u (i, j)] of size ( km) × (dn).

Linear convolution has the form:

The hardware implementation of the spatial filter filter core is shown in FIG.

Block 1 of the reference frame in the form of a matrix of photosensitive elements (photodiodes) has a size kxd. The signal from each photosensitive element is amplified by a separate amplifier 2 (operational amplifiers A1-An) with a gain of Ku, which is necessary to increase the signal level and stable filtering in the P-center and in the M-periphery, as well as the stable operation of the adders in the frequency band F. The amplified signals are summed with various coefficients, which are set by resistors 4 (R1.1-R1.n). Weighting coefficients are selected based on the values of the wavelet function of the core at the corresponding points for adder 3, forming the P-center, and resistors 6 (R2.1-R2.n) for adder 5, forming the M-periphery. The difference between the P-center and the M-periphery with the subsequent rise of the output signal to the positive region generated by the algebraic adder 7 is the resulting useful signal, on the basis of which the image U is formed.

The original images and the resulting after convolution by the core of the spatial filter, as well as their histograms are presented in figure 5.

Thus, the claimed method of organizing the core of spatial filtering and its subsequent linear convolution with images in different spectral ranges allow you to normalize the image. The histograms of normalized images are centered and tend to one- or two-mode look. Such normalized images are well binarized, while the binarization threshold is selected from a whole range of values, thereby achieving a pronounced robust character, which ensures an acceptable quality of binarization in a wide range of changing background noise conditions.

Therefore, due to the introduction of an additional arsenal of technical means - operations for the method and elements and blocks for the device - the corresponding technical results are achieved - expanding the scope of the method and expanding the functionality of the device.

Claims (2)

1. The method of organizing the core of spatial filtering, based on the sequential formation of fragments of the original image in the form of discrete matrixes of video signals, characterized in that they form a binary image of the fragments of the original image by their initial linear convolution with a spatial filter, the core of which contains the central and peripheral parts, while the initial linear convolution is performed separately with the central and linear parts of the core of the spatial filter, from the results of which the modules are formed x difference used as a binary image of the original image fragments, the central part of the spatial filter has an exponential form, and the peripheral portion has negative coefficients and generated using exponential basis of the condition that the sum of the central and peripheral parts. 2. The device for implementing the method according to claim 1, containing a block of the reference frame in the form of a discrete matrix of video signals, as well as filtering means, characterized in that a group of amplifiers is introduced, the input of each of which is connected to the corresponding output of the block of the reference frame, and the filtering means contain the central and peripheral parts, each of which is made in the form of an adder with weighted inputs, and an algebraic adder, while the inputs of the adders with weighted inputs are connected to the outputs of the amplifiers, and the outputs are connected to the inputs s algebraic combiner module output signal which is the output device.

Images