RU2755092C1 - Method for forming image with local brightness gradient and device for its implementation - Google Patents

Abstract

FIELD: computer technology.

SUBSTANCE: invention relates to computer technology. A method for forming an image with a local brightness gradient is proposed, in which phase distortions of the object image are artificially formed, images of the object corresponding to its various phase distortions are recorded, the registered images are digitized and processed, while during processing, a sample average and a sample brightness variance from the obtained images are calculated for each pixel, then the ratio of the sample variance to the sample average is calculated for each pixel and it is raised to a given degree, the resulting value is put in accordance with this pixel and the resulting image of the object is built.

EFFECT: formation of an image with a local brightness gradient, which is resistant to the uneven transmission coefficient of the optical system of the photodetector.

2 cl, 5 dwg

Description

The invention relates to the field of television technology, in particular, to systems for processing and forming images and can be used to detect, classify or evaluate the parameters of objects.

The most common methods for generating images with a local luminance gradient are methods based on spatial filtering of images. In these methods, a simple difference filter or more complex integrating-differentiating filters that are resistant to the noise of a photodetector are used as a filter. At the same time, the most optimal is the luminance filter with the kernel in the form of the derivative of the Gaussian function (Canny J. "A Computational Approach to Edge Detections IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. PAMI-8 no. 6, 1986, pp. 679- 698).

The known method and device for highlighting the contours of image objects (patent for invention of the Russian Federation No. 2383925, IPC G06K 9/50, publ. 10.03.2010, BI No. 7). The method consists in spatial filtering of the image by the Gaussian function and the subsequent application of the Sobel operator.

The known method and device for detecting local features in the image (patent for invention of the Russian Federation No. 2535184, IPC G06T7 / 40, publ. 10.12.2014, BI No. 34). The method is based on the method of spatial filtering of an image with kernels of various types, allowing not only to determine the presence of a gradient, but also the type of local feature that caused it (ripples, spots, waves, edges).

The known method, which consists in the fact that taking into account the level of image noise, select the smoothing coefficient, calculate the coefficients of smoothing cubic B-splines along each row and each column of the image. Then, two matrices and a component of the brightness gradient are formed at each point of the image by summing the smoothing parabolic B-splines with the previously found coefficients of the smoothing cubic B-splines for rows and columns. Next, the modulus of the brightness gradient is determined at each point of the image. After that, the contours of the object are formed by transforming the modulus of the brightness gradient, in the process of which, on a new white matrix, elements are highlighted in black, the modulus of the gradient for which in the coordinates of the image exceeds the transformation threshold (patent for invention of the Russian Federation No. 2695417, IPC G06K 9/48, publ. 23.07 .2019, BI No. 21).

The general disadvantages of the above methods are the high computational complexity of the spatial filtering operation and the lack of consideration of the dependence of the quality of the resulting image on the non-uniformity of the transmission coefficient of the optical system of the photodetector. Uneven transmittance can be caused both by the lens design (so-called vignetting) and by various small-scale optical irregularities (dust, scratches, streaks) on the surface or in the thickness of the protective glass of the photodetector matrix. Small-scale irregularities have the greatest effect in long-focus systems and can lead to the formation of parasitic images of diffraction rings or dark spots in the vicinity of the irregularity. As a result of the spatial filtering of such an image, in addition to the useful signal, a false signal is also formed, caused by the difference in brightness in the region of small-scale inhomogeneity. Vignetting of the image, as a rule, manifests itself in the form of a low-frequency multiplicative component of the signal and, therefore, can lead to a false overestimation of the gradient value obtained by the methods of spatial filtering of the image.

A known method for automatic correction of image artifacts caused by small-scale irregularities, in particular - dust particles on the glass of the photodetector (US patent No. 2015110417, IPC G06T 5/00, H04N 5/217, publ. 23.04.2015). In this method, areas of the image corresponding to artifacts are determined, after which the image is corrected based on the samples of neighboring image pixels. The disadvantages of this method are the high complexity of the calculations and the dependence of the quality of the resulting image on the reliability of the used computation model to real artifacts.

The closest in technical essence is the method (prototype) according to US patent No. 2015103181, IPC H04N 5/357, G06T 5/00, H04N 5/33, publ. 04/16/2015). The essence of the prototype method lies in the automatic correction of the unevenness of the transmittance of the optical system over the image field by digital image processing. For this, the prototype proposes to form two images: the first is the original, the second is calculated from the first by applying low-frequency digital spatial filtering. The resulting image is formed by element-wise dividing the first image into the second.

Essential features of the prototype:

- compensation of the low-frequency component of uneven brightness;

- the use of two images to form the result;

- formation of the resulting image by element-wise division of one image into another;

- the use of digital spatial filtering of the image.

The disadvantage of the prototype is the high sensitivity of the processed resulting image to small-scale components of brightness inhomogeneities. Another disadvantage of the prototype is that object boundaries are significantly distorted as a result of object image processing.

The purpose of the invention is the formation of an image with a local brightness gradient, which is resistant to uneven transmittance of the optical system of the photodetector.

This goal is achieved by the fact that, according to the claimed method, phase distortions of the object image are artificially formed, the object images corresponding to its various phase distortions are recorded, the registered images are digitized and processed, while the processing for each pixel of the images calculates the sample mean and sample variance of the brightness from of the obtained images, after that, the first resulting image is formed, in which each of its pixels is associated with the calculated value of the sample variance, the second resulting image is formed, in which the calculated value of the sample average is assigned to each of its pixels, then the ratio of the sample variance to the sample mean and raise it to a given power, on the basis of the two resulting images, an output image of the object under study is formed, in which each of its pixels is assigned to This is the calculated value of the ratio of the sample variance to the sample mean, raised to a given power.

A device that implements the method comprises a variable phase screen, an optical lens, a matrix photodetector, an analog-to-digital converter and a signal processing unit, connected in series with each other, the variable phase screen being formed by a gas flow blown at a predetermined flow rate.

The essence of the claimed method is illustrated in FIG. 1-5. FIG. 1 shows the sequence of image processing of the object under study, where 1 is the object, 2 is a sequence of images with different phase distortions, 3 is the result of calculating the sample variance in each pixel of the image sequence, 4 is the result of calculating the average sample value in each pixel of the image sequence, 5 is the local the brightness gradient of the image (the result of the implementation of the method). FIG. 2 shows a block diagram of a device for implementing the proposed method, where 6 is a variable phase screen, 7 is an optical system lens, 8 is a multi-element photodetector, 9 is an ADC, 10 is a signal processing unit. FIG. 3 shows a photograph of a fragment of a circular light mark with uneven transmission of the optical system. FIG. 4 shows the image of the local brightness gradient obtained by the linear filtering method with the kernel of the form of the derivative of the Gaussian function. FIG. 5 shows an image with a local brightness gradient obtained using the claimed method.

The essence of the proposed method lies in the fact that it is proposed to form a certain predetermined sequence of images of the same object with different phase distortions, and then perform special digital processing of the resulting sequence of images. Digital processing consists in calculating for each pixel of the output image the ratio of the sample variance to the average sample value of the brightness for a given pixel of the generated sequence of images. The result of this digital processing is an image with a local brightness gradient.

Features similar to the prototype are:

- compensation of the low-frequency component of uneven brightness;

- using two images to get the result;

- formation of the result by element-wise division of one image into another.

The method includes the following sequence of actions:

- artificially form phase distortions of the object image, thereby changing the brightness distribution of the object image;

- register a sequence of images with different phase distortions with a multi-element photodetector;

- digitize images using an analog-to-digital converter (ADC);

- calculate the sample variance and sample mean value of the brightness in each pixel of the image;

- calculate the ratio of the sample variance to the sample mean in each pixel of the image, the result of the calculation is taken as the output signal of the brightness in this pixel.

The method is based on the relationship between the brightness variance and its average value: for a constant luminous flux, the number of signal carriers n accumulated in one pixel of the photodetector during the accumulation time is subject to Poisson statistics. In particular, the variance of the number of signal carriers D (n) is equal to its mathematical expectation E (n).

For image reading systems with a linear or approximately linear light-signal characteristic, the digital signal S at the ADC output of the image reading system linearly depends on the number of signal carriers:

where

K is the end-to-end transmission coefficient of the image reader.

In this case, the variance of the samples of the digital signal S at the ADC output, without taking into account the effect of the quantization effect, is determined by the ratio:

From relation (1) it follows that at a constant luminous flux, the ratio of the variance of the luminance signal at the ADC output to its mathematical expectation is a constant value:

For areas of the image with a zero brightness gradient, the result of calculating D (S) / E (n) is a constant value, and it is determined by expression (2). Otherwise the D (S) / E (n) value will be the larger, the larger the module of the image gradient. The amplitude of phase distortions introduced during the formation of the sequence of images determines the degree of blurring of the output signal between adjacent pixels of the image, and, therefore, the degree of detail in the output image. To obtain clearer boundaries (degree of detail) of the generated image, the operation of raising the ratio D (S) / E (n) to a given power N is used. In this case, the greater the degree of elevation N, the higher the clarity of the image, but the lower its contrast and brightness. Therefore, the value of the degree is determined for each object individually and usually amounts to N = 1-2.

The stability of the object image to uneven transmittance of the optical system, incl. caused by small-scale irregularities in the glass of the photodetector and vignetting is provided by the invariance of expression (2) to any multiplicative stationary component of the optical image. Thus, the method does not require preliminary calibration for the non-uniformity of the transmittance of the optical system.

The image formed according to the claimed method is also resistant to two types of photodetector noises, namely, to a random noise component and to the noise of the dark current of the photodetector matrix. Resistance to the random noise component of the photodetector is provided due to the accumulation of statistics (the number of images) necessary to calculate the sample variance and mean value. Resistance to inhomogeneity of the dark current of the pixels of the photodetector is provided by the invariance of expression (2) to the value of the dark current. The invariance is ensured due to the fact that the number of signal carriers accumulated due to the thermal generation process, as well as accumulated due to photogeneration, obeys the Poisson distribution.

An additional advantage of the claimed invention is the relative ease of image processing. Since each pixel of the image is processed independently and only requires the calculation of the ratio D (S) / E (n), therefore, image processing can be implemented using relatively simple computational tools, and potentially directly on the photodetector, which makes it possible to reduce the number of elements of the image processing system.

One of the possible options for the artificial formation of phase distortions of the object image is the use of a weakly turbulent or time-varying laminar gas flow, which can be used as an intermediate medium between the object and the optical system of the photodetector. It is known that the optical refractive index of a gaseous medium depends on its density. In turbulent and changing laminar flows, a time-varying density gradient occurs. As a consequence, the refractive index also changes over time and, thus, the gaseous medium can play the role of a time-varying phase screen.

A device that implements the inventive method consists of sequentially connected with each other optoelectronic elements: a variable phase screen 6, an optical lens 7, a matrix photodetector 8, an analog-to-digital converter 9 and a signal processing unit 10 (Fig. 2). In this case, the role of the variable phase screen is played by the gas flow blown at a given flow rate.

The device works and is used as follows. With the help of the optical lens 7, the image of the object 1 is focused on the matrix photodetector 8, while using the variable phase screen 6 one or another phase distortion of the image of the object 1 is set. When optical radiation from the object 1 hits the matrix photodetector 8, a signal corresponding brightness, i.e. a matrix of brightness values is formed. The indicated brightness values are digitized in the ADC 9 and fed to the signal processing unit, in which the calculated algorithm for the brightness of the image pixels is carried out according to the claimed method. As a result of the device operation, an image with a local brightness gradient is formed.

The claimed method and device make it possible to form a high-quality image with a local gradient of image brightness, which, in turn, makes it possible to determine the contours of the objects under study for their subsequent detection, classification or assessment of their parameters.

Claims (2)

1. A method of forming an image with a local brightness gradient, which consists in the fact that phase distortions of the object image are artificially formed, the object images corresponding to its various phase distortions are recorded, the registered images are digitized and processed, while during processing, a sample is calculated for each pixel of the images. the average and sample variance of brightness from the obtained images, then the first resulting image is formed, in which each of its pixels is associated with the calculated value of the sample variance, the second resulting image is formed, in which the calculated value of the sample average is assigned to each of its pixels, then for each pixel, the ratio of the sample variance to the sample mean is calculated and raised to a given power, on the basis of the two resulting images, an output image of the object under study is formed, in which each its pixel is associated with the calculated value of the ratio of the sample variance to the sample mean, raised to a given power. 2. A device for forming an image with a local brightness gradient, which implements the method according to claim 1, which comprises a variable phase screen, an optical lens, a matrix photodetector, an analog-to-digital converter and a signal processing unit connected in series with each other, wherein the variable phase screen is formed gas flow blown at a given flow rate.

Images