RU2728949C1 - Method of constructing and processing images and system for implementing thereof - Google Patents

Abstract

FIELD: data processing.

SUBSTANCE: invention relates to construction and processing of primary RAW image into digital form. Disclosed is a method of constructing and processing images on a plurality of colours in a Bayer arrangement for an optical imaging system using an image sensor having a plurality of elements of an optical system, in form of pixels to form an unprocessed degraded image of object (RAW), wherein each pixel of the image sensor for each part of the image data of the corresponding plurality of colours in the Bayer layout is coated with one of a plurality of colour filters, comprising steps of dividing, RAW image scanning - R - G - B colours, each pixel colour processing, as if it were image data of one colour, stage of recursion of scanning of processed image data, stage of restoration of each pixel of colours for image in Bayer layout.

EFFECT: technical result consists in improvement of quality of procedure for construction and processing of dynamic image.

2 cl, 8 dwg

Description

The invention relates to the field of construction and processing of a primary RAW image in digital form (compression format), fully adapted for all processing classes: storage and conversion (lossy compression, batch compression, editing); analysis (recognition, filtering); synthesis (approximation, vectorization), in the formats of static images with losses, as well as any moving images, which is implemented in the form of a basic device for creating an end-to-end presentation format and processing digital images in any signal range, depending on existing scanning tools. To substantiate the technical solutions adopted in solving these problems, let us briefly consider the theoretical foundations of the prior art for the construction and processing of images.

State of the art

Currently, when forming an image model, the mathematical classical Gauss-Markov linear regression model is widely used with a data structure with BLUE estimates (best linear unbiased estimates) (consistent, unbiased and effective estimates) and with an additive model of visual noise accompanying the formation of images. This model describes well the effect of film grain, fluctuation noise in radio engineering systems, quantization noise in analog-to-digital converters, etc. There are many methods of image processing that reduce noise and significantly improve their quality, but when noise suppression occurs, loss or distortion of data. This leads to a loss of resolution ¹ ( ¹ Resolution of the digital camera matrix - the ability of the device to convey fine details of the image detailing the image. The resolution of a graphic image is determined by the number of pixels horizontally and vertically per unit of length (inch) of the image. The smaller the dot size , the higher the resolution (more raster lines and dots per line) and, accordingly, the higher the image quality. The value of the resolution is usually expressed in dpi (dot per inch - dots per inch), i.e. in the number of dots in a strip of image length one inch (1 inch = 2.54 cm) (loss of fine image details) and degrades their detail.

Algorithms for suppressing noise in images, implemented on computing devices, require the formation of a method for optimal scanning of a two-dimensional image, since the algorithm is, by definition, a sequence of some steps with linearly addressable computer memory, and also due to the sequential method of transmitting messages and organizing the storage of such data in the computer memory for performing transformations on array elements. This requirement determines the need to solve the problem of V.V. Alexandrova [Alexandrov R.V. Representation and processing of images. Recursive approach / R.V. Alexandrov, I. D. Gorsky. L .: Nauka, 1985. 102 pp.], Which consists in the following: for a given class of images, find representations of a two-dimensional array R ² in the form of a linearly ordered sequence of elements R ¹ , taking into account the multidimensional connectivity of the elements of the topology of the multidimensional image space and the linearity of the computer architecture. It should be borne in mind that the cost of solving the problem grows exponentially with increasing dimension.

I. Formation of scanning (sweep). When scanning an image, information on all state variables can be represented by an equivalent one-dimensional time series in m-dimensional phase space using the method of time delays according to Takens F. Detecting Strange At-tractors in Turbulence. - Berlin, Springer, 1981. - P. 366-381]

where m is the dimension of the embedding, τ is the time delay.

In this case, this time series x _i (t), where t = i τ is a set of observations {x _i } = [x _1, x ₂ , ..., x _N ], drawn through a discrete time interval τ. The interval τ can be constant and requires a uniform scale of observations. And vice versa, the interactions of states and their number in the phase trajectory, which preserves the structures of the original phase trajectory, is reconstructed from the time series embedded in the pseudophase space of a given dimension m:

в m - мерной фазовой траектории состояний системы x(t) длиной N на двумерную квадратную двоичную матрицу размером m×m, в которой единица (черная точка) соответствует повторению состояния при некотором времени i в некоторое другое время j, а обе координатные оси являются осями времени (рекуррентная диаграмма (recurrence plot, RP)) для фиксации информации о рекуррентном поведении системы.In 1987, Eckmann (Eckmann) and co-authors [Eckmann J. - P., Kamphorst SO, Ruel-le D. Recurrence Plots of Dynamical Systems // Europhysics Letters. 1987. 5. P. 973-977] proposed a way to display the trajectory

in the m - dimensional phase trajectory of the states of the system x (t) of length N into a two-dimensional square binary matrix of size m × m, in which the unit (black point) corresponds to the repetition of the state at some time i at some other time j, and both coordinate axes are the axes time (recurrence plot, RP) to capture information about the recurrent behavior of the system.

Thus, the problem arises of choosing the type of scanning of a two-dimensional array in one-dimensional and setting the order on the set of elements of a one-dimensional data array. The choice of the type of scanning is determined by a compromise between the efficiency of maintaining the mutual topological proximity of elements of the original and expanded spaces, the property of quasi-continuity and the preservation of correlations between elements in both spaces, as well as the simplicity of the scanning algorithm.

Currently, the generally accepted method of two-dimensional scanning is line-by-line (columnar) scanning of the information field, also called television scanning. In this scan, the structural links that exist between the blocks of the original image are broken. This approach is based on abandoning line scanning of a digital image and replacing it with scanning a more complex type. In [Alatar A.I., Mikhailov A.A. Algorithms for scanning (scanning) information systems for the restoration of dynamic images // Bulletin of the Tula State University. Technical science. - 2018. Issue. 10. - P. 399-411] shows the basic types of scanning, the simplest linear scanning by rows (columns), as well as spiral scanning, recursive scanning (Hilbert scan).

In an effective way of ordering the elements of an array, an algorithm is used that traverses the pixels sequentially, grouping them [RV Alexandrov. Representation and processing of images. Recursive approach / R.V. Alexandrov, I. D. Gorsky. L .: Nauka, 1985.102 p .; Sagan N. Space-Filling Curves / H. Sagan. - New York: Springer-Verlag. 1994. 208p] and forming a one-dimensional sequence, which is convenient to store and transmit via communication channels to select individual objects (since the points belonging to it are grouped). Currently, line scanning of an image using self-similar mappings according to the Curve Fill Plane (FFC) rules is widely used. Displaying an image through self-similar squares with a side of 3 ^p , where p is the degree of convergence, p = 0 ... 10, into the processor memory is performed according to the rules of the QPC or segments of length 9 ^p . The proposed scanning of image elements (pixels) allows (due to memory organization and over-parallel access) to increase the speed of image processing by tens of times for all its classes of procedures. This KZP, unlike others, has been worked out right up to its hardware implementation [Warren G., Jr. Algorithmic tricks for programmers. "Williams", Moscow, 2004.288 p. p. 233]. However, such methods are not effective enough, since they do not take into account the characteristic features of images as objects of processing: the two-dimensional nature of statistical relationships, the presence of homogeneous areas, contours, etc. Taking into account the specified specifics makes it possible to improve the quality of processing, but leads to the complication of the corresponding algorithms. Moreover, it is possible to take into account the features of images not at the stage of data processing, but at the stage of their formation.

II. Selection and justification of the optimality of scanning (sweep) when building an image.

To date, many scanning algorithms have been proposed, based on the concept of QPC [Aleksandrov R.V. Representation and processing of images. Recursive approach / R.V. Alexandrov, I. D. Gorsky. L .: Nauka, 1985.102 p .; Sagan N. Space-Filling Curves / H. Sagan. - New York: Springer-Verlag. 1994.208p .; Valantinas J. On the use of space-filling curves in changing image dimensionality / J. Valantinas // Information technology and control. 2005. Vol. 34. no. 4. P. 345-354], which do not take into account the features of the images. These shortcomings of the known algorithms for removing noise in the image stimulate new approaches to the tasks of their preprocessing, which are able to preserve this important characteristic of the image processing algorithm. The most important characteristic of the quality of a raster image is the resolution, which determines the spatial sampling of continuous images, the criterion for which is the criterion resolution - measurement [Shirman Y.D. Resolution and Compression of Signals / Ya.D. Shirman M .: Sov. radio. 1974. 360 p.], Taking into account the random nature of the resulting image.

The choice of the scanning method and the corresponding scan, which, on the one hand, are simply implemented, and on the other, lead to acceptable loss of information about the image properties, accompanying the transition from a two-dimensional image to its one-dimensional data vector, seems relevant. The possibility of solving this problem is determined by the Takens theorem [Broer N. Dynamical Systems and Chaos / H. Broer, F. Takens. New York: Springer. 2011. 313 pp], which defines the transition from a two-dimensional image to its one-dimensional data vector using the immersion of time series in the phase space of the original image.

One of the ways to improve the quality of an image in the presence of noise arising in the process of its formation is to increase its detail by choosing the type of scanning (scanning) of a two-dimensional array in a one-dimensional sequence, taking into account the multidimensional connectivity of elements and the linearity of the computer architecture, i.e. optimization of scanning algorithms (sweep) by immersion in the phase space of the image.

The QPC is based on the Lipschitz mapping, the mapping ƒ. X → Y between two metric spaces, the application of which increases the distance by no more than some constant times, i.e. a mapping ƒ of a metric space (X, ρ _X ) into a metric space (Y, ρ _Y ) is Lipschitz if there is a Hölder-Lipschitz constant L of this mapping (the Hölder-Lipschitz condition

for any x :, y∈X.

состоянию

квазинепрерывного рекурсивного сканирования Гильберта-Пеано (СГП) [Александров Р.В. Представление и обработка изображений. Рекурсивный подход/Р.В. Александров, И.Д. Горский. Л.: Наука, 1985. 102 с.] используют рекуррентность сканирования в виде окрестностного индекса элемента (ОИЭ), который определяется как1. Formation of indicators of scanning topology. When ordering a two-dimensional array R ² on R ^{1 by a} specific scan in an ordered one-dimensional set, subsets G are determined, consisting only of elements of the two-dimensional neighborhood of the element under consideration, one of which includes this element itself. The number of elements of this "one-dimensional" subset of G characterizes the property of preserving a specific neighborhood of a given scan. In recursive scanning, the smallest structural unit is the reference cell (generating element) of G elements, the number of elements in which is called the neighborhood index of the element under consideration (OIE) J _e [Alexandrov R.V. Representation and processing of images. Recursive approach / R.V. Alexandrov, I. D. Gorsky. L .: Nauka, 1985. 102 s]. The local characteristic J _e is a function of the scanning law and the location of the element in the converted array. To analyze the topological proximity of the state

condition

quasi-continuous recursive scanning Hilbert-Peano (SGP) [Alexandrov R.V. Representation and processing of images. Recursive approach / R.V. Alexandrov, I. D. Gorsky. L .: Nauka, 1985. 102 pp.] Use the recurrence of scanning in the form of a neighborhood index of an element (OIE), which is defined as

, i,j=1…N, N - количество рассматриваемых состояний x_i, ε_i - размер окрестности точки

в момент i,

норма и Θ(⋅) - функция Хэвисайда:Where

, i, j = 1 ... N, N is the number of considered states x _i , ε _i is the size of the neighborhood of the point

at time i,

norm and Θ (⋅) is the Heaviside function:

, попадающие в m - мерную окрестность с радиусом ε_i и центром в

.Recurrent are the states of the recurrence points

falling into an m-dimensional neighborhood with radius ε _i and center at

...

. Если точка x_j попадает внутрь данной окрестности, то такое состояние считается подобным состоянию

. Радиус ε_i выбирается либо непосредственно для каждой точки, исходя из получения в такой окрестности некоторого заранее определенного количества рекуррентных данной точек, либо постоянным, что дает симметричную диаграмму и является наиболее используемым вариантом.The size of the neighborhood ε _i determines the radius of the neighborhood in the phase space centered at the point

... If a point x _j falls into a given neighborhood, then such a state is considered similar to the state

... The radius ε _i is chosen either directly for each point, based on obtaining in such a neighborhood a certain predetermined number of recurrent given points, or constant, which gives a symmetric diagram and is the most used option.

оказывает влияние на вид диаграммы. Как правило, используются нормы L₁, L₂ (евклидова норма), L_∞ (максимальная норма): норма L_∞ обеспечивает нахождение наибольшего (фиг. 1), L₂ - среднего (фиг. 2), a L₁ - минимального количества соседних точек (фиг. 3). При построении рекуррентных диаграмм используется норма L_∞, так как она независима от размерности фазового пространства, наиболее просто вычисляема и позволяет изучать диаграммы теоретически, поскольку аналитические выражения для L_∞ решаются проще, нежели для любой другой нормы [Thiel М. et all. Influence of observational noise on the recurrence quantification analysis//Physica D. 2002. 171 (3). P. 138-152].Norm type

affects the appearance of the chart. As a rule, the norms L ₁ , L ₂ (Euclidean norm), L _∞ (maximum norm) are used: the L _∞ norm ensures finding the maximum (Fig. 1), L ₂ - the average (Fig. 2), and L ₁ - the minimum neighboring points (Fig. 3). When constructing recurrent diagrams, the L _∞ norm is used, since it is independent of the dimension of the phase space, it is most easily calculated and allows you to study diagrams theoretically, since analytical expressions for L _∞ are easier to solve than for any other norm [Thiel M. et all. Influence of observational noise on the recurrence quantification analysis // Physica D. 2002.171 (3). P. 138-152].

In addition to depicting recurrence in the form of black dots, it is possible to depict the distance between the states of the system on the distance diagram [Iwanski J. S., Bredley E. Recurrence plots of experimental data: To embed or not to embed? // Chaos - 1998. - No. 8 (4). - P. 861-871], which is displayed on a certain color palette:

When they are used in a one-dimensional signal, the two-dimensionality of the statistical relationships of image elements manifests itself mainly in the form of an increase in the correlation interval, which makes it possible to more efficiently apply simple algorithms for processing one-dimensional data immersed in the phase space.

To characterize the scan as a whole, use the average value of the RIE over N = k ^pm (where k is the number of parts into which the side of the hypercube (image) of dimension p is divided for each subsequent partition of m) to the array elements

which is called the neighborhood scan index J _c (OIS) and is a function of the scan law and reflects its properties.

Different two-dimensional scans have different properties of preserving the "neighborhood" of the elements of the i array, and the range of changes in the OIS is in the range of 2.0-4.6 [N. Gorsky, S.N. Mysko, V.P. Sukharichev. Comparative study of some characteristics of two-dimensional sweeps. Preprint LNIVTS AN SSSR, No. 44, L., 1982. p. 10], which can be divided into three parts. The simplest scans occupy the lower part of the range with values of the neighborhood scan index J _c (∞) = 2. Different types of recursive scanning (k = 2, 3) with different types of reference cells occupy the upper part of the range with the corresponding values of the OIS J _c . An intermediate position is occupied by recursive scans that do not have the quasi-continuity property.

Since the OIE is a function of the scanning law and in a particular scan depends on the location of the array element, an unsuccessful choice of the type of scan can lead to the destruction of information contained in the relative position of the image elements. This defines the task of choosing from a variety of scanning methods that introduce less distortion in the desired field.

The most important scanning properties are continuity and block structure. Continuity of scanning is expressed in the fact that adjacent samples in the sequence are adjacent in the image. In this case, the scanning trajectory is "held" for a long time in each local area of the image, providing a strong correlation of the sequence of samples, and the block structure makes it possible to carry out recursive scanning algorithms [Alexandrov R.V. Representation and processing of images. Recursive approach / R.V. Alexandrov, I. D. Gorsky. L .: Nauka, 1985.102 p.].

The indicators of the topological characteristics of the OIS and OES scanning are not sufficient for choosing a scanning method, since they are not oriented towards use in image processing and are functionally redundant for this application (and, therefore, complex). The choice of the scanning method is determined not only by its topology, but also by the class of images and the purpose of their processing. From the point of view of their capabilities, image scanning methods should retain fractal and statistical properties when they are converted to one-dimensional form. This will allow you to choose a scanning method that is compromise in terms of its implementation complexity and preservation of the fractal properties of the image. In this regard, let us compare scanning properties by statistical analysis of an ordered two-dimensional array, converted by a specific scan to an ordered one-dimensional form.

2. Preservation of locality in homogeneous Hilbert curves is determined by the curve itself, that is, the used affine transformations and the chosen kernel. An important characteristic of Peano curves, closely related to the Hölder-Lipschitz condition [4], is the square-to-linear ratio of dilation ² ( ² Dilatation (from Latin dilatatio) denotes expansion, dilution, stretching.) [Bauman K.E. Peano-Hilbert Curve Stretching Coefficient // Mathematical Notes. T. 80 Issue. 5 November 2006.S. 643-656; Shchepin E.V., Bauman K.E. Minimal Peano Curve // Trudy Mathematical Institute. V.A. Steklova, 2008, V. 263, pp. 251-271], which for a pair p (t), p (τ) of points ³ ( ^{3 A} point of a curve is a point of its graph, that is, a point of a Peano curve is a pair of t, p (t), where t belongs to the segment being mapped, p (t) is the square-image.) Peano ZPK p: [0, 1] → [0, 1] × [0, 1] the value

.

...

The upper bound of the square-to-linear ratio for all possible pairs of different points on the curve is called the square-to-linear ratio of the curve. Curves with the smallest square-linear ratio are of interest for applications, for which the local preserving properties of the mapping are better [Shchepin EV, Bauman K.E. Minimal Peano Curve // Trudy Mathematical Institute. V.A. Steklova, 2008, T. 263, S. 251-271].

, меньшим чем у кривой Пеано-Гильберта, глубина⁴ (⁴ Глубиной правильной фрактальной кривой Пеано называется наибольшее натуральное число d, для которого кривая имеет стык порядка d, не подобный никакому стыку меньшего порядка. Ограничение кривой Пеано на пару соседних фрактальных периодов порядка k называется стыком порядка к этой кривой.) которой не превосходит 9.Among the Peano curves of fractal genus 9 by Theorem 8 [Shchepin EV, Bauman K.E. Minimal Peano Curve // Trudy Mathematical Institute. V.A. Steklov, 2008, T. 263, pp. 251-271], a unique minimal N-shaped Peano curve (Figs. 4, 5) with a square-linear ratio

less than that of the Peano-Hilbert curve, depth ⁴ ( ^{4 The} depth of a regular fractal Peano curve is the largest natural number d for which the curve has a junction of order d, which is not similar to any junction of a lower order. The restriction of a Peano curve to a pair of adjacent fractal periods of order k is called a junction of order to this curve) which does not exceed 9.

3. Selection and justification of the optimality of image processing

Correct fractal Peano curve ⁵ ( ⁵ Correct fractal Peano curve is a mapping of a segment into a square, which allows partitioning the domain of definition into several equal segments (fractal periods) such that the limitation of the curve to any of its fractal periods is similar to the entire curve. The number (minimum possible) of fractal periods is called the fractal genus of the curve.) p (t), which maps a unit segment to a unit square (unit curve) is parameterized by the area, i.e. the area of the image of any interval is equal to the length of this interval.

BDS-статистикой [Шаповалов А.В. Выбор метода развертки цифровых изображений при анализе их фрактальных признаков в фазовом пространстве с использованием BDS-теста/А.В. Шаповалов //Системи

та

К.: 2012. Вип. 3 (23). С. 116-121; Kanzler Ludwig. Very Fast and Correctly Sized Estimation of the BDS Statistic/Ludwig Kanzler//Christ Church and Department of Economics University of Oxford. - 1999. - 95 с], погруженному в m-мерное псевдофазовое пространство, осуществляют путем параметризации площадью кросс-рекуррентной диаграммы изображения. Все типы сканирования с разбросом ОИС от единицы до четырех, в отсутствии шума, дают надежные результаты локализации искусственных объектов на изображении по вычисленному полю BDS-статистик при нелинейном анализе временных рядов.Quantitative analysis by parametrization of the scanning (sweep) area by immersion in the phase space of the image [Alatar A.I., Mikhailov A.A. Algorithms for scanning (scanning) information systems for the restoration of dynamic images // Bulletin of the Tula State University. Technical science. - 2018. Issue. 10. - P. 399-411] in the selection of fractal features of images by the method of time series analysis

BDS statistics [Shapovalov A.V. Choice of a method for scanning digital images when analyzing their fractal features in phase space using the BDS test / A.V. Shapovalov // System

that

К .: 2012. Vip. 3 (23). S. 116-121; Kanzler Ludwig. Very Fast and Correctly Sized Estimation of the BDS Statistic / Ludwig Kanzler // Christ Church and Department of Economics University of Oxford. - 1999. - 95 s], immersed in the m-dimensional pseudo-phase space, is carried out by parameterization by the area of the cross-recurrent diagram of the image. All types of scanning with a scatter of OIS from one to four, in the absence of noise, give reliable results of localization of artificial objects in the image using the calculated field of BDS statistics in nonlinear time series analysis.

, формируется гистограмма распределения BDS-статистики

(аббревиатура от фамилий ее предложивших) [Brock W.A. A test for independence based on the correlation dimension/Brock W.A., Dechert W.D., LeBaron В., Scheinkman J.A.//Econometric Reviews. 1996. №15(3). P. 197-235]. Анализ BDS-статистик, полученных для различного сканирования, позволяет с заданной достоверностью принимать гипотезу о том, что сканирование, для которой BDS-статистика меньше, в большей степени разрушает зависимости между элементами (пикселями) в изображении.As a result of the BDS test, when scanning an image, the elements of which are expanded into a vector

, a histogram of the BDS statistics distribution is formed

(an abbreviation from the names of those who proposed it) [Brock WA A test for independence based on the correlation dimension / Brock WA, Dechert WD, LeBaron V., Scheinkman JA // Econometric Reviews. 1996. No. 15 (3). P. 197-235]. The analysis of BDS statistics obtained for various scans makes it possible to accept the hypothesis that scanning, for which the BDS statistics is smaller, to a greater extent destroys the relationships between elements (pixels) in the image.

, погруженному в m-мерное псевдофазовое пространство, величиной

:BDS statistics [Brock WA A test for independence based on the correlation dimension / Brock WA, Dechert WD, LeBaron V., Scheinkman JA // Econometric Reviews. 1996. No. 15 (3). P. 197-235] is determined by the calculation of the time series

immersed in an m-dimensional pseudophase space of size

:

where σ _{m, n} (ε) is the standard deviation of the random variable C _{m, N} (ε) - [С _{1, N} (ε)] ^m .

Then, for a sample of N observations, the correlation integrals C _{m, N} (ε), C _{1, N} (ε) for the phase space of dimension m are calculated using the following expression:

- функция Хэвисайда:Where

- Heaviside function:

и

, заданных своими проекциями (x_s, x_s+1, …, x_s+m) и (x_t, x_t+1, …, x_t+m) в m-мерном фазовом пространстве, в m-мерную гиперсферу радиуса ε, а знаменатель равен среднеквадратическому отклонению σ_m,N(ε) числителя.determining the frequency of hitting an arbitrary pair of points

and

given by their projections (x _s , x _{s + 1} , ..., x _{s + m} ) and (x _t , x _{t + 1} , ..., x _{t + m} ) in the m-dimensional phase space, into the m-dimensional hypersphere of radius ε, and the denominator is equal to the standard deviation σ _{m, N} (ε) of the numerator.

. Зависимость корреляционного интеграла от ε имеет степенной вид

где D_c - корреляционная размерность временного ряда. Корреляционная размерность определяется средним числом точек х_j от х_i, расстояние между которыми меньше εBDS statistics are based on the use of the correlation dimension (correlation integral) of a process in a pseudophase space of a given dimension m. The correlation integral С _m determines the frequency of hitting an arbitrary pair of points in the phase space in the vicinity of the ε-radius of the hypersphere, which tends to a certain limit (fractal dimension) as ε decreases. The BDS statistics are calculated for processes immersed in a pseudophase space of dimension m. The value of the BDS statistics tends to a certain limit as ε decreases in the range ε = 0.5σ ÷ 2σ, where σ is the standard deviation of the series

... The dependence of the correlation integral on ε has a power-law form

where D _c is the correlation dimension of the time series. Correlation dimension is determined by the average number of points x _j from x _i , the distance between which is less than ε

- вектор в n-мерном фазовом пространствеwhere С (ε, N) is the correlation integral; ε is the correlation radius; N is the number of points used to estimate the dimension; Θ - Heaviside function;

- vector in n-dimensional phase space

со 100% вероятностью при N→∞, a

является случайной величиной с асимптотически гауссовским распределением с нулевым средним и среднеквадратическим отклонением

, которое определяется как:The value of C _{m, N} depends on the dimension of the pseudophase space. Brock et al. Showed [Brock WA A test for independence based on the correlation dimension / Brock WA, Dechert WD, LeBaron B., Scheinkman JA // Econometric Reviews. 1996. No. 15 (3). P. 197-235; Schreiber T. Discrimination power of measures for nonlinearity in a time series / A. Schmitz // Physical Review E. - 1997. - V. 55, No. 5. - P. 5] that

with 100% probability as N → ∞, a

is a random variable with an asymptotically Gaussian distribution with zero mean and standard deviation

which is defined as:

Where

которой близка к ее теоретическому значению σ(ε) [Сосулин Ю.Г. Фрактальное обнаружение протяженных малоконтрастных объектов на изображениях/Ю.Г. Сосулин, А.Б. Русский// Радиотехника. 2009. №9. 12. С. 48-57].For processes with independent values, identically distributed (IID-independent and identical distributed) random variables, BDS-statistics have a normal distribution law with zero mean, the estimate

which is close to its theoretical value σ (ε) [Sosulin Yu.G. Fractal detection of extended low-contrast objects in images / Yu.G. Sosulin, A.B. Russian // Radio engineering. 2009. No. 9. 12. S. 48-57].

III. Separation of multiple colors in an image. RAW image signal processing is carried out according to one of the R-, G- and B- color components, which is formed by dividing the light flux with a color filter into spectral components - red (R), green (G) and blue (B) - according to the number of image converters on the video matrix. The color components of the corresponding pixels composing the RAW image are arranged according to the Bayer arrangement, the frequency characteristics in the G-component pixel arrangement in the RAW image differ from the frequency characteristics in the R- and B-component pixel arrangement. Since the image recovery processing is equivalent to frequency response correction, the G-component bandwidth is different from the R-component or B-component bandwidth.

If the input image is an RGB color image, it is preferable to generate three image reconstruction filters corresponding to the respective color components, i.e. to read the R-, G- and B-components of the image pixel by pixel, for example, a mask of red 110000, green 001100 and cyan 000011 can be applied to each pixel.

IV. Continuous wavelet transform (CWT) is an integral transform, which is the convolution of ψ (t) - the mother wavelet function (mother wavelet) with the signal x (t), defined on the time axis of the variable t [Addison P.S. The Illustrated Wavelet Transform Handbook. - IOP, 2002. P. 351]:

- весовая функция.where ψ ^* denotes complex conjugation for ψ, the parameter b∈R corresponds to the time shift, and is called the position parameter, the parameter a > 0 specifies the scaling and is called the stretching parameter,

- weighting function.

The wavelet transform converts the signal from a time representation to a time-frequency representation, which allows you to compress the original data set. Wavelets (English wavelet "small wave") are local in time and frequency and are obtained from one basic one, changing it (shifting, stretching).

фурье-преобразование для вейвлета ψ(t), то естьIf a

the Fourier transform for the wavelet ψ (t), that is

then the admissibility condition must be satisfied:

или вейвлет ψ(t) должен иметь среднее равное нулю.those. the wavelet with a zero frequency component must satisfy the condition

or the wavelet ψ (t) must have a mean equal to zero.

The original function is restored by the inverse transformation

In numerical and functional analysis, discrete wavelet transforms (DWT) refer to wavelet transforms in which wavelets are represented by discrete signals (samples).

The claimed method uses the simplest wavelet transform in the form of a step function (Fig. 6)

those. Haar transform (PX) [Kazaryan M.L. Investigation of Haar Wavelet Transforms for Correctness in the Context of the Earth Space Monitoring Problem. North Caucasian region. Natural Sciences. 2013. No. 6. Pp. 14-17], whose Laplace image with the Laplace transform parameter p is equal to

V. Image restoration. The degraded image g (x, y) and the original image ƒ (x, y) are related through the point spread function (PSF) h (x, y) obtained by the inverse Fourier transform of the optical transfer function (OTF) ⁶ ( ⁶ OTF (optical transfer function) is obtained by Fourier transform for PSF.):

g (x, y) = h (x, y) * ƒ (x, y),

where * represents convolution and (x, y) represents coordinates in the image.

When this equation is converted to a display on the frequency plane by means of a Fourier transform, it takes the form of a product for each frequency, as represented in the equation:

where H (u, ν) is the Fourier transform for PSF h (x, y) and represents the OTF, and F (u, ν) is the Fourier transform for the degraded image ƒ (x, y), represent the coordinates on the two-dimensional frequency plane, i.e. .e. frequency.

To obtain the original degraded image from the read raw (RAW) image, both sides of equation (1) can be divided by H (u, ν), as represented in the equation,

where R (u, ν) is called the image reconstruction filter.

The actual image includes noise components. For this reason, using an image reconstruction filter formed by taking the ideal inverse function from the OTF in the above manner will amplify the noise components along with the degraded image. Currently, a Wiener filter is used to suppress noise in an image.

Since OTF varies when imaging in terms of image scale and aperture diameter, it is necessary to change the image reconstruction filter used for image reconstruction processing accordingly.

Performing image recovery processing on a readable input image can improve image quality by correcting aberrations.

Analogs and their disadvantages

есть квадрат со стороной 3^р, где р=0…N, при этом N задает сторону квадрата, в который вкладывают габарит исходного изображения; при р=0 (0Fas) точка вырождается в пиксел; при р>0 МЦР разбивает плоскость на

независимых точек (квадратов); параметры х, у для Q - это его центр на плоскости; в результате образуют многокоординатное пространство точки Q: две координаты декартового измерения (х, у), координата измерения вращения (w), координата измерения размера граней фасета pFas, координата измерения номера хранения

в памяти вычислителя F^p; причем эти координаты измерений комбинируют в зависимости от поставленных задач; общим для этих координат является отображение декартовой плоскости в память при взаимно однозначном и непрерывном отображении точки

в память вычислителя F^p.A known method of converting and processing a digital image based on multi-center scanning [Patent No. 2517715 RF G06T 5/00, H04N 7/00 Publ. 05/27/2014, Bul. No. 15], built according to the rules of the curve filling the plane (CPF), the initial cell of the multicentric sweep (MCR) is represented by a discrete square, consisting of nine cells (3 × 3 = 9), which has its center and its four faces (sides); the scanning of the initial cell of the MCR is performed from the center to the edge of the square and then bypassing the remaining cells in a circle; the priority is the path with the bypass direction to the left from the center of the square and further clockwise; the generated construction is a facet (pFas), where p is the recursion step, for p = 1, they have the above-described initial cell (3 × 3 = 9); To construct further directions of recursion, four types of traversal are distinguished: the above traversal w1 as initial (1Fas1), traversal w2 as mirror from 1Fas1 to the left (1Fas2), traversal w3 as mirror from 1Fas2 to the top (1Fas3), traversal w4 as mirror from 1Fas3 to the right; to obtain the direction of the MCB recursions, 2Fas is used (p = 2, with a side of 9 cells, 9 × 9 = 81), where 1Fas is set as the center; on the basis of the above rotations of w, a tour is performed in a sequence with an initial square movement to the left of 1Fas and then clockwise around 1Fas: w1 w2 w3 w4 w3 w2 w3 w4 w3; then the MCB recursion is started based on 1Fas and 2Fas (p> 2), representing the plane in rotation coordinates (w), the carrier of which is pFas, and the point of this plane,

there is a square with a side of 3 ^p , where p = 0 ... N, while N sets the side of the square into which the size of the original image is inserted; at p = 0 (0Fas) the point degenerates into a pixel; for p> 0, the MCR divides the plane into

independent points (squares); parameters x, y for Q are its center on the plane; as a result, a multi-coordinate space of the point Q is formed: two coordinates of the Cartesian dimension (x, y), the coordinate of the rotation measurement (w), the coordinate of the dimension of the facet facet pFas, the coordinate of the measurement of the storage number

in the memory of the calculator F ^p ; moreover, these coordinates of measurements are combined depending on the tasks; common to these coordinates is the mapping of the Cartesian plane into memory with a one-to-one and continuous mapping of the point

into the memory of the calculator F ^p .

The disadvantages of this method are

- blurring of image details regardless of the type of noise, which characterizes their resolution, due to the low topological connectivity of pixels of the used Gilbert-Peano curve during scanning (scanning);

- the complexity of the hardware implementation, since algorithms for suppressing noise in images, implemented on computing devices, require the formation of a method for optimal scanning of a two-dimensional array R ^{2 of an} image in the form of a linearly ordered sequence of elements R ¹ , in order to reduce the cost of solving the problem, determined exponentially from its dimension ;

- a low degree of relevance (i.e., in terms of information value for the observer) at the initial level of image presentation due to the allocation of semantic areas of interest in the image without preliminary processing;

- low performance due to the increased size of the convergence area;

- low compression ratio of the image;

- low efficiency of work with digitized signals of any frequency range and received by any type of sensors.

Known device alphabetic image compression format without loss [Utility model to patent No. 138747 RF H04N 1/047, H04N 19/189 Publ. 03/20/2014 Bul. No. 8], in which the input image enters the input image decomposition unit from a raster in an area of the square type using self-similar mappings from the center of the square in the form of a multicentric sweep (MCR) to obtain, based on the enumerated alphabet, the form of a text file with a meaningful representation of image elements for the observer , the output of which is connected to the input of the block for generating the output text file-package from the denoting, symbolic and conceptual components of the image, presented in a relevant form, distinguishing: chaos, compacts, contours, stand-alone lines; directly on the resulting form, as a compression format, further all classes of image processing, including direct approximation and vectorization, are performed.

The disadvantages of this device are

- blurring of image details regardless of the type of noise, which characterizes their resolution, due to the low topological connectivity of pixels of the used Gilbert-Peano curve during scanning (scanning);

- the complexity of the hardware implementation, since the algorithms for suppressing noise in images implemented on computing devices require the formation of a method for optimal scanning of a two-dimensional array R ^{2 of an} image in the form of a linearly ordered sequence of elements R ¹ , to reduce the cost of solving the problem, determined exponentially from its dimension ;

- a low degree of relevance (i.e., in terms of information value for the observer) at the initial level of image presentation due to the allocation of semantic areas of interest in the image without preliminary processing;

- low performance due to the increased size of the convergence area;

- low compression ratio of the image;

- low efficiency of work with digitized signals of any frequency range and received by any type of sensors.

Prototypes and their flaws

The closest technical solution to the invention (prototype) is a method of image processing [Patent No. 2523924 G06T 5/00, H04N 5/357 Publ. 07/27/2014 Bul. No. 21], in which to perform image data recovery processing to correct image deterioration due to aberration in the optical imaging system, said image data obtained by reading an image of an object passed through the optical imaging system using an image sensor having a plurality of pixels, with each pixel of the image sensor being covered by one of a plurality of multi-color color filters, the method comprising:

- a separation step in which the image data of a plurality of colors of a plurality of color filters is divided into a plurality of image data of corresponding colors, and in the dividing step, image data of one color whose spatial frequency response is higher than that of another color due to the arrangement of the color filters of the plurality of colors are divided into a plurality of image data of said one color, so that a plurality of image data of said one color each has the same spatial frequency characteristic as image data of a different color;

- an image processing step in which reconstruction processing is performed by processing with a filter for each of the image data of the respective colors separated in the separation step using a plurality of respective image processing means arranged sequentially, wherein in the image processing step, each of the plurality of image processing means performs reconstruction processing on image data of each color that has not been subjected to reconstruction processing, and separately treats image data of one color whose spatial frequency response is higher than that of another color due to the arrangement of color filters of a plurality of colors as data image of a plurality of colors that is treated separately so that a portion of the image data that is treated separately as a plurality of colors and which is processed by a plurality of image processing means has the same the same frequency response as image data of a different color;

- an interpolation processing step in which color interpolation processing of each pixel is performed on the image data subjected to the reconstruction processing by using the image data of a plurality of colors as if it were image data of one color.

The disadvantages of this method are

- blurring of image details, regardless of the type of noise, which characterizes their resolution, due to the low topological connectivity of pixels during scanning (scanning);

- the complexity of the hardware implementation, since algorithms for suppressing noise in images, implemented on computing devices, require the formation of a method for optimal scanning of a two-dimensional array R ^{2 of an} image in the form of a linearly ordered sequence of elements R ¹ , in order to reduce the cost of solving the problem, determined exponentially from its dimension ;

- a low degree of relevance (i.e., in terms of information value for the observer) at the initial level of image presentation due to the allocation of semantic areas of interest in the image without preliminary processing;

- low performance due to the increased size of the convergence area;

- low compression ratio of the image;

- low efficiency of work with digitized signals of any frequency range and received by any type of sensors.

The closest technical solution to the invention (prototype) is an image processing device [Patent No. 2523924 G06T 5/00, H04N 5/357 Publ. 07/27/2014 Bul. No. 21], in which to perform image data recovery processing to correct image deterioration due to aberration in the optical imaging system, said image data obtained by reading an image of an object passed through the optical imaging system using an image sensor having a plurality of pixels, with each pixel of the image sensor being covered by one of a plurality of color filters, the image processing apparatus comprising:

- separating means for separating image data of a plurality of color filter colors into image data of corresponding color filter colors in a Bayer arrangement, which is configured to separate image data of one color whose spatial frequency response is higher than that of another color due to the arrangement of the plurality of color filters of a plurality of colors on a plurality of image data of said one color, so that the plurality of image data of said one color has the same spatial frequency response as image data of a different color;

- a plurality of sequential image processing means, each of which is designed to perform recovery processing by filter processing image data of one of the respective colors separated by said separation means, which is configured to perform convolution processing for said filter, and to perform data recovery processing image to correct image deterioration due to aberration in the optical imaging system, said image data being obtained by reading an image of an object passed through the optical imaging system using an image sensor having a plurality of pixels, with each pixel of the image sensor being covered by one of a plurality of color filters, each of said plurality of image processing means is configured to perform recovery processing for a corresponding their image data of each color that has not been subjected to reconstruction processing, and separately treats image data of one color whose spatial frequency response is higher than that of another color due to the arrangement of color filters of multiple colors, as with image data of multiple colors, so that a portion of the image data that needs to be treated separately as image data of a plurality of colors and which is to be processed by said plurality of image processing means has the same frequency response as image data of a different color, and for processing with a filter, performed by a plurality of image processing means, an appropriate filter is used, comprising a two-dimensional filter obtained by an inverse Fourier transform for a function formed on the basis of an inverse function of an optical transfer function of an optical element of an optical imaging system niy, and the above-mentioned plurality of image processing means is further configured to perform convolution processing for the above-mentioned filter;

- interpolation processing means for performing interpolation processing of the colors of each pixel as if it were image data of the same color undergoing reconstruction processing, wherein the filter used for filter processing performed by said image processing means comprises a two-dimensional filter obtained by performing the inverse Fourier transform for the function generated based on the inverse function of the transfer function of the optical element of the optical imaging system.

The disadvantages of this device are

- blurring of image details, regardless of the type of noise, which characterizes their resolution, due to the low topological connectivity of pixels during scanning (scanning);

- the complexity of the hardware implementation, since algorithms for suppressing noise in images, implemented on computing devices, require the formation of a method for optimal scanning of a two-dimensional array R ^{2 of an} image in the form of a linearly ordered sequence of elements R ¹ , in order to reduce the cost of solving the problem, determined exponentially from its dimension ;

- a low degree of relevance (i.e., in terms of information value for the observer) at the initial level of image presentation due to the allocation of semantic areas of interest in the image without preliminary processing;

- low performance due to the increased size of the convergence area;

- low compression ratio of the image;

- low efficiency of work with digitized signals of any frequency range and received by any type of sensors.

Description of the invention.

The technical problem solved by the claimed method is to improve the quality of the procedure for constructing and processing a dynamic image. A method of converting and processing a digital image based on a minimum N-shaped scan (MNC) (minimum N-shaped scan (MNP)) in the form of a Peano minimum N-shaped diagonal curve with a 9th fractal genus is intended for converting a line (raster) representation of an image into self-similar squares according to mappings of the QPC type. The use of MNC (MNP) allows several times to increase the efficiency of the implementation of procedures for the existing classes of the image processing system.

Comparative analysis with the prototype shows that the technical result of the proposed device is determined by:

- reduction of blurring of image details regardless of the type of noise, which characterizes their resolution during scanning (scanning);

- reducing the complexity of the hardware implementation and the cost of solving the problem, since the complexity of the problem grows exponentially with increasing dimension;

- increasing the degree of relevance (i.e., in terms of information value for the observer) at the initial level of image presentation, since the algorithms for suppressing noise in images, implemented on computing devices, take into account the multidimensional connectivity of the elements of the multidimensional image space and the linearity of the computer architecture;

- an increase in performance, which is achieved by reducing the area of convergence.

- increasing the degree of image compression;

- the ability to efficiently work with digitized signals of any frequency range and received by any type of sensors.

The method has software and hardware implementation.

The technical result is achieved by the fact that to solve these problems in the proposed method for constructing and processing images on a plurality of colors in the Bayer arrangement for an optical imaging system using an image sensor having a plurality of elements of the optical system, in the form of pixels to form a raw degraded image of an object (RAW ), wherein each pixel of the image sensor for each piece of image data of the corresponding set of colors in the Bayer arrangement is covered by one of a variety of color filters, including:

a separation step in which the image data of a plurality of colors are divided by a plurality of color filters into a plurality of images of corresponding —R — G — B colors;

- an image processing step in which reconstruction with a filter is sequentially performed for each part of the image data of the corresponding plurality of colors in the Bayer arrangement, separated in the separation step, using a plurality of appropriate image processing means, after which the image processing is corrected, and further includes:

- the stage of scanning RAW images - R - G - B colors, passed through the optical imaging system based on the minimum N-shaped scan (MNC);

the step of processing the colors of each pixel for the image data subjected to the reconstruction processing is performed by using the image data of a plurality of colors as if it were image data of the same color;

- a step of recursively scanning the processed image data based on minimum N-scan (MNP);

- the stage of restoration of each pixel of colors for the image in the Bayer arrangement;

- at the initial stage, when implementing the minimum N-shaped scanning (MNC), a unit square-image (K ₀ 1) with side 1 of the processed image is divided into nine congruent ⁷ ( ⁷ Congruence (lat.congruens - proportional, corresponding, coinciding) of two figures or two parts of one figure, if one of them can be translated into another by shifting, rotating and mirroring (or their composition) sub-squares and the corresponding unit segment [0; 1] (O ₀ 1) is also divided into nine equal segments and the sub-squares are numbered sequentially (K ₁ 1; K ₁ 2; K ₁ 3; K ₁ 4, ..., K ₁ 9), K ₁ 1 neighbor K ₁ 2, K ₁ 2 neighbor K ₁ 3 and K ₁ 3 neighbor K ₁ 4, etc. and a segment into 9 identical parts (segments O ₁ 1, O ₁ 2, O ₁ 3, O ₁ 4, ..., O ₁ 9), the beginning of the curve is selected in the lower left corner, and the end is in the upper right and determines which segment into which sub-square falls (in this case, adjacent segments pass into sub-squares adjacent on the side), for each small segment, coordinates m are determined points of the sub-square, into which its beginning, end and middle fall, then each small segment is divided into nine parts and similar constructions are carried out, etc. until the size of congruent sub-squares is approximately equal to the size of the pixel, the congruent sub-squares are determined equal to the pixel when forming the minimum initial cell of the unit square-image, scanning at the zero step of the initial cell (recursion direction) starts along the minimum N-shaped diagonal Peano curve with the 9th fractal genus, from the beginning in the lower left corner of the unit square, the beginning of the segment, to the upper right corner of the unit square-image, the end of the segment, at the first step, along the path with the direction of traversing up from the beginning of the unit square-image and further up, then turning to the right, down, down, right, up, up to the end of the unit square-image; the formed initial cell (3 × 3 = 9) at p = 1 is facet ⁸ ( ⁸ Facet is a beveled face of something.) (pFas), where p is the recursion step;

- scanning that implements the minimum N-shaped scanning (MNC), RAW images of the object are carried out cyclically along a discrete square facet, built on the basis of scanning according to the rules of the curve filling the plane (CPF), with the setting of the beginning and direction of recursion, in the form of a minimum N-shaped diagonal a Peano curve with a 9th fractal genus, for which the initial scanning cell (square-image) is set by a discrete initial unit square-image consisting of nine cells (3 × 3 = 9) and having four faces (sides), the beginning and end of which are located at opposite vertices of the unit square-image;

- bypassing the sub-squares with minimal N-shaped scanning (MNC) is performed by transitions through the rotation operators (w) from one to the other according to the principle of side proximity, and the transition through the sub-square is carried out through its center, then a one-to-one correspondence is established between the segments and sub-squares, so that adjacent segments pass through the rotation operators (w) into adjacent (on the side) sub-squares, i.e. O ₁ 1 in K ₁ 1, O ₁ 2 in K ₁ 2, etc., to construct further directions of recursion, three types of rotation operators (w) for traversing are distinguished: the above rotation operator w1 upward as the initial one (1Fas1) from the beginning unit square-image, the rotation operator w2 as a mirror from 1Fas1 to the right (1Fas2), the rotation operator w3 as a mirror from 1Fas2 down (1Fas3), and to obtain the subsequent direction of recursions, 2Fas is used (p = 2, with a side of 9 cells, 9 × 9 = 81), where the start specifies the end of 1Fas, etc .;

- on the basis of the above operators of rotations w, they traverse in a sequence with the initial movement to a sub-square up from 1Fas and further: start → w1 → w1 → w2 → w3 → w3 → w2 → w1 → w1 → end, and the formation of the scanning curve is carried out by connecting the centers sub-squares with a line in the order of traversal, obtained on the second division, entering the sub-square K ₁ 1 in the lower left corner, and ending in the upper right corner, a Peano curve is built by sequentially traversing all the sub-squares obtained in the previous division, which remains the same (K ₁ 1 → K ₁ 2 → K ₁ 3 → K ₁ 4 →… → K ₁ 9) and redefines the location of the images of the points of the segment and the traversal rule in each sub-square K _n i up to a rotation similar to the traversal rule in the original unit square K ₀ 1, defining the Peano curve as a fractal ⁹ ( ^{9 A} self-similar geometric figure is a figure that can be cut into a finite number of identical figures similar to itself. Objects with such a property, American mathematician Benoit Mandelbrot pr He proposed to be called fractals (from lat. frangere - "break", "break") .;

- at the processing stage, a wavelet transform is carried out in the form of a Haar transform over the signal value of each pixel from the scanned set of image states of the image sensor, the value of each pixel obtained by the Haar transform is fed to image processing for reconstruction using a filter in the form of a filter for solving the inverse problem from the optical the transfer function of an element of the optical system for forming RAW images, after which the image processing is corrected, and at the end of the image processing, the inverse Haar transform is performed over the set of images,

согласованное со сканированием по закону MNP, представляя плоскость в координатах вращения (w), носителем которых является фасет - квадрат pFas с точкой этой плоскости

, имеющего свой центр (ху), со стороной 3^p и площадью 9^р, где р=0…N, при этом N задает сторону единичного квадрата, в который вкладывают габарит исходного изображения; при р=0 (0Fas) точка вырождается в пиксель;- at the stage of the minimum N-shaped recursive sweep (MNP), the value obtained restored after the inverse wavelet transform is stored in the memory block of the calculator according to the measurement coordinate of the storage number

consistent with scanning according to the MNP law, representing the plane in rotation coordinates (w), the carrier of which is the facet - square pFas with the point of this plane

, having its center (xy), with a side of 3 ^p and an area of 9 ^p , where p = 0 ... N, while N sets the side of the unit square in which the size of the original image is inserted; at p = 0 (0Fas) the point degenerates into a pixel;

независимых точек (подквадратов);for p> 0, the plane is divided into

independent points (sub-squares);

parameters x, y for Q are its center on the plane;

в памяти вычислителя F^p;as a result, the multi-coordinate space of the point Q is formed: two coordinates of the Cartesian dimension (x, y), the coordinate of the rotation measurement (w), the coordinate of the dimension of the facet facet pFas, the coordinate of the measurement of the storage number

in the memory of the calculator F ^p ;

в память вычислителя F^p, причем переход на последующий фасет сканирования RAW изображения происходит после полного заполнения соответствующего фасета развертки блока в памяти вычислителя F^p, а окончание восстановления изображения осуществляется после заполнения всей площади развертки изображения.common to these coordinates is the mapping of the Cartesian plane into memory with a one-to-one and continuous mapping of the point

into the memory of the calculator F ^p , and the transition to the next facet of the RAW image scanning occurs after the corresponding facet of the block scan in the memory of the calculator F ^{p is} completely filled, and the end of image recovery is carried out after filling the entire scan area of the image.

An image construction and processing system that implements the proposed method comprises an imaging unit using an image sensor having a plurality of pixels, wherein each pixel of the image sensor is covered with one of a plurality of color filters forming a separation unit for separating the image data of a plurality of colors of color filters into image data with analog memory devices at the output corresponding to the colors of the color filters in the Bayer arrangement, a signal processing channel containing a signal processing unit, which is implemented on the corresponding filter obtained on the basis of the inverse function of the convolution of the filtered signal with a kernel in the form of an optical transfer function of an optical element of the formation system images, the signal correction unit, the input of which is connected to the output of the signal processing unit, and the output to the output of the signal processing channel, and the display unit, the outputs of the color separation unit are connected to the inputs of the scanning unit image processing, the information output of which is connected to the information input of the signal processing channel, and the image scanning unit consists of multiplexer units and the address setting unit, for scanning the RAW image along the diagonal minimum N-shaped Peano curve, while maintaining the original topological and correlation connectivity of the two-dimensional image, an input for the start of scanning was additionally introduced into the system, and a wavelet transform unit was additionally introduced into the signal processing channel, using the example of a Harr transform unit, as well as an inverse Haar transform unit, a scanning unit containing an analog demultiplexer unit and an analog memory unit, and a display unit, in addition, ten signal delay elements and an "OR" element, and the address inputs of the multiplexer unit are connected to the outputs of the address setting unit, the information output of the image scanning unit is connected to the information input of the signal processing channel, which consists of from sequentially located Harr conversion unit, signal processing unit, the information output of which is connected to the information input of the signal correction unit, the output of which through the signal processing channel output is connected to the information input of the inverse Haar conversion unit, the information output of the latter is connected to the input of the scanner, the input of which is connected with the information input of the demultiplexer unit, the address inputs of which are connected to the information outputs of the block for setting the addresses of the image scanning unit, and the information outputs of the demultiplexer unit are connected to the information inputs of the analog memory unit and then through the outputs of the scanner with the inputs of the display unit, the input of the first signal delay element is connected with recording inputs of output analog storage devices (AAM) of color filters of the color separation unit, and the output of the first signal delay element is connected to the input of the installation in the initial state of the block back The output of the multiplexing end of the image scanning unit is connected to the timing input of the Haar transform unit and through the third signal delay element to the timing input of the signal processing unit and the input of the fourth delay element signal, the output of which is connected to the clock input of the signal correction unit and through the fifth signal delay element to the inverse Haar transform clock input of the inverse Haar transform unit and then through the sixth delay element is connected to the clock input of the demultiplexer unit, the output of the sixth delay element is also connected through the seventh delay element with the clock input of the analog memory block of the scanner and then through the tenth signal delay element with the counting input of the block for setting the addresses of the image scanning unit, and through the eighth element signal delay with the clock input of the display unit and then through the ninth signal delay element with the first input of the "OR" element, the second input of which is connected to the input of the system scan start, and the output with the input of the first signal delay element.

FIG. 1, 2, 3 show the norms equal to the radius around the point of the phase space (black point), depicted for the two-dimensional phase space: Fig. 1 - L ₁ fig. 2 - L ₂ , fig. 3 - L _∞ , in Fig. 4 is the minimum diagonal ¹⁰ ( ¹⁰ Peano curves, the beginning and end of which are located at opposite vertices of the square-image, are called diagonal.) An N-shaped curve of the 9th kind (MNC curve) at p = 1, in Fig. 5 is the minimal diagonal N-shaped curve of the 9th kind (curve MNC) for p = 2, in Fig. 6 scaling ¹¹ ( ¹¹ Scale invariance or scaling is a property of the equations of physics, to keep its form when all distances and time intervals change in the same number of times.) - Haar function, in Fig. 7 shows a block diagram of a system for constructing and processing a dynamic image (SPiOI), and FIG. 8 shows an example of the implementation of the address setting block (BZA), in which the code of the MNC curve is formed on the basis of the binary code.

The block diagram (Fig. 7) of the system for constructing and processing an image (SPiOI) contains an imaging unit (BFI) 1 using an image sensor (ID) 2 having a plurality of pixels, and each pixel of the image sensor is covered with one of a plurality of color filters (CF ) 3 forming a separation unit (BR) 4 for separating the image data of a plurality of colors of color filters (CF) 3 into image data with analogue storage devices (CAM) at the output corresponding to the colors of the color filters (CF) 3 in the Bayer arrangement, an image scanning unit (BSI) 5, which consists of multiplexer units (BMUX) 6 and an address setting unit (BZA) 7, for scanning a RAW image along the diagonal minimum N-shaped Peano curve, while maintaining the original topological and correlation connectivity of a two-dimensional image, scan start input 8 , the signal processing channel (KOS) 9 in the form of sequentially located wavelet transform blocks, for example, the block Harr transform (BPH) 10, signal processing unit (BFB) 11, and the signal processing unit 11 is implemented on the corresponding filters obtained on the basis of the inverse function of the convolution of the filtered signal with the kernel in the form of an optical transfer function of the optical element of the imaging system, and the said unit image processing KOS 9 is additionally configured to carry out convolution processing for the said filter, a signal correction unit (BCS) 12, an inverse Haar transform unit (BOPH) 13, a scanning unit (BR) 14, containing a block of analog demultiplexers (BDMUX) 15 and an analog block storage devices (BASE) 16, as well as a display unit (BO) 17, ten signal delay elements τ ₁ 18 - and τ ₁₀ 27 and an OR element 28.

Examples of the implementation of analog memory devices (ADS) can be found in the book Tietze W., Schenck K. Semiconductor Circuitry: A Reference Guide. Per. with him. M .: Mir, 1982. - 512 p., For example, implementation in S. 284 fig. 17.17.

Examples of the implementation of analog multiplexers MUX (DMUX) (demultiplexers) are given in the work of Pukhalsky G.I., Novoseltseva T.Ya. Digital devices: a textbook for technical colleges. - SPb .: Polytechnic, 1996 .-- 885 p. on pp. 461-472, for example, the implementation in fig. 6.55.

In SPiOI (Fig. 7), the outputs of the color separation unit (BRC) 4 are connected to the inputs of the image scanning unit (BSI) 5 to scan RAW through the multiplexer unit (BMUX) 6 of the image along the diagonal minimum N-shaped Peano curve, while maintaining the original topological and correlation connectivity of a two-dimensional image. The address inputs of the multiplexer unit (BMUX) 6 are connected to the outputs of the address setting unit (BZA) 7. The information output of the image scanning unit (BSI) 5 is connected to the information input of the signal processing channel (COS) 9. The signal processing channel (COS) 9 consists of a series located Harr transformation unit (BPH) 10, signal processing unit (BFB) 11, and the signal processing unit (BFB) 11 is implemented on the corresponding filters obtained on the basis of the inverse function of the convolution of the filtered signal with the kernel in the form of an optical transfer function of the optical element of the formation system images. The information output of the signal processing unit (BFB) 11 is connected to the information input of the signal correction unit (BCS) 12, the output of which is through the output of the signal processing channel (COS) 9 is connected to the information input of the Haar inverse transformation unit (BOPH) 13. Information output of the inverse conversion unit Haar (BOPH) 13 is connected to the input of the scanner (BR) 14, the input of which is connected to the information input of the demultiplexer unit (BDMUX) 15, the address inputs of which are connected to the information outputs of the address setting unit (BZA) 7 of the image scanning unit (BSI) 5, and the information outputs of the demultiplexer unit (BDMUX) 15 are connected to the information inputs of the analog storage unit (BASE) 16 and then through the outputs of the scanner unit (BR) 14 with the inputs of the display unit (BO) 17.

The input of the first signal delay element τ ₁ 18 is connected to the recording inputs of the output analog storage devices (ADS) of the color filters 3 of the color separation unit (BRC) 4, and the output of the first signal delay element τ ₁ 18 is connected to the input of the installation in the initial state of the address setting unit ( BZA) 7, the counting input of which through the second signal delay element τ ₂ 19 is connected to the clock input of the multiplexer unit (BMUX) 6 of the image scanning unit (BSI) 5. The output of the end of multiplexing (BMUX) 6 of the image scanning unit (BSI) 5 is connected to the input clocking the Haar transform unit (BPH) 10 and through the third signal delay element τ ₃ 20 with the clock input of the signal processing unit (BFB) 11 and the input of the fourth signal delay element τ ₄ 21. The output of the fourth signal delay element τ ₄ 21 is connected to the clock input of the block signal correction (BCS) 12 and through the fifth signal delay element τ ₅ 22 with the clock input of the inverse Haar transform of the block about rat Haar transform (BOPH) 13 and then through the sixth delay element τ ₆ 23 is connected to the clock input of the demultiplexer unit (BDMX) 15. The output of the sixth delay element τ ₆ 23 is also connected through the seventh delay element τ ₇ 24 to the clock input of the analog memory block (BASE) 16 of the scanner (BR) 14 and then through the tenth signal delay element τ ₁₀ 27 with the counting input of the address setting unit 7 of the image scanning unit (BSI) 5, and through the eighth signal delay element τ ₈ 25 with the clock input of the display unit ( BO) 17 and further through the ninth signal delay element τ ₉ 26 with the first input of the OR element 28. The second input of the OR element 28 is connected to the input of the start of scanning 8 of the system, and the output to the input of the first signal delay element τ ₁ 18.

Let us consider the implementation of the method using the example of the work of SPiOI (Fig. 7).

, имеющего свой центр (ху), со стороной 3^р и площадью 9_р, где p=0…N, при этом N задает сторону единичного квадрата, в который вкладывают габарит исходного изображения. Такое "измельчание" изображения продолжается до тех пор пока DBS - статистика асимптотически не сведется к гауссовскому распределению с нулевым средним и соответствующим среднеквадратическим отклонением.A matrix of addresses is preliminarily formed for scanning a RAW image along the diagonal minimum N-shaped Peano curve, which is placed in the address setting unit (BZA) 7. For this, the image is "refined" by representing it by a plane in rotation coordinates (w), the carrier of which is a facet - square pFas as a diagonal minimal N-shaped Peano curve with a point on this plane

, having its center (xy), with a side of 3 ^p and an area of 9 _p , where p = 0 ... N, while N sets the side of the unit square in which the size of the original image is inserted. This "refinement" of the image continues until the DBS - statistics asymptotically reduce to a Gaussian distribution with zero mean and the corresponding standard deviation.

The imaging unit (IMU) 1 uses image sensors (ID) 2 having a plurality of pixels, with each pixel of the image sensor being covered by one of a plurality of color filters (CF) 3 forming a color separation unit (BRC) 4 to separate a plurality of image colors color filters 3 for R-, G- and B-components in the Bayer arrangement, for subsequent processing for each component separately. The signal from the corresponding output of the separation unit 4 is scanned by the image scanning unit (BSI) 5 according to the law of the diagonal minimum N-shaped Peano curve (Fig. 4, 5) with the L _∞ norm (maximum norm), which ensures finding the largest number of neighboring points (Fig. 3 ). At the same time, the original topological and correlation connectivity of the two-dimensional image is preserved.

The address inputs of the multiplexer unit (BMUX) 6 are connected to the outputs of the address setting unit (BZA) 7. On command at the start of scanning input 8 of the image scanning unit (BSI) 5, the address setting unit (BZA) 7 is set to its initial state. The signal from the output of the multiplexer unit (BMUX) 6 of the image scanning unit (BSI) 5 is fed to the input of the signal processing channel (COS) 9. At the input of the signal processing channel (COS) 9, its input signal is converted into a signal-image by the Haar transform unit (BPH) 10. Further, the obtained signal-image of the Haar transform is processed by the signal processing unit (BFB) 11, and the signal processing unit (BFB) 11 is implemented on the corresponding filter obtained on the basis of the inverse function of the convolution of the filtered signal with the kernel in the form of an optical transfer function of the optical element of the system imaging. The signal from the output of the signal processing unit (BFB) 11 of the signal processing channel (KOS) 9 enters the signal correction unit (BKS) 12 and then to the inverse Haar transform unit (BOPH) 13, where it is converted into the original signal by the signal of the end of signal correction with the output of the signal correction unit (BCS) 12. The reconstructed signal is fed to the scan by the scanning unit (BR) 14 through the demodulator unit (BDMUX) 15 at the address set at the address input from the output of the address setting unit (BZA) 7, and arrives at the information input for writing to the block of analog storage devices (BASE) 16 and then to the display unit (BO) 17, where a joint image in the Bayer arrangement is formed from the colors selected for processing with color filters (CF) 3 of the color separation unit (BRC) 4.

In order to synchronize the sequence of processing the information signal from the processed image, strobe signals for each signal processing block are formed on the signal delay elements in the SPIE. All elements of the signal delay τ ₁ 18 - τ ₁₀ 27 can be combined into a single control unit and are chosen equal to the response time of the unit to the input of which they are connected, for reliable operation of this unit and reliable establishment of the signal at its output. So the recording signal of analog storage devices (ADS) of color filters (CF) 3 through the first signal delay element τ ₁ 18 sets the address setting unit (BZA) 7 to the initial scanning state, and the counting input of the address setting unit (BZA) 7 through the second delay element signal τ ₂ 19 strobes the multiplexer unit (BMUX) 6. The signal from the output of the end of switching of the multiplexer unit BMUX 6 strobes the Haar transform unit 10 and through the third signal delay element τ ₃ 20 strobes the signal processing unit (BFB) 11 and then through the fourth signal delay element τ ₄ 21 strobes the signal processing unit (BCS) 12. The signal from the output of the fourth signal delay element τ ₄ 21 through the fifth signal delay element τ ₅ 21 strobes the inverse Haar transform unit (BOPH) 13 and then through the sixth signal delay element τ ₆ 23 strobes at the gating input, the operation of the demultiplexer unit (BDMUX) 15. The signal from the output of the sixth signal delay element τ ₆ 23 after the delay n and the seventh delay element τ ₇ 24 records the signal from the output of the demultiplexer unit (BDMUX) 15 at the write input to the analog storage unit (BASE) 16 of the scanning unit (BR) 14. The signal from the output of the seventh delay element τ ₇ 24 through the tenth signal delay element τ ₁₀ 27 is fed to the counting input of the address setting unit (BZA) 7 and transfers to the next address for scanning the RAW signal through the BMUX 6 multiplexer unit the images along the diagonal minimum N-shaped Peano curve, as well as the address of the scan by the scanning unit (BR) 14. Through the eighth signal delay element τ ₈ 25 strobes the display unit (BO) 17 and through the ninth signal delay element τ ₉ 26 and the first input of the OR element 28, this signal writes the output signals of the color filters at the write input to the output analog storage devices (AMU) (DF) 3. The initial start of the system for constructing and processing images is carried out at the input of the beginning of scanning 8 through the second input of the OR element 28.

Thus, MNC creates a multi-dimensional point Q space:

- two coordinates x, y,

- rotation coordinate (w),

is the nesting coordinate of a point in the multi-coordinate space Q,

- coordinate of the facet size of the pFas facet,

- coordinate of embeddings of the processed point from O ^p-1 to Q ^p ,

в памяти вычислителя F^p.- storage address

in the memory of the calculator F ^p .

что определяет возможность использования при фильтрации сканированного ряда фильтра Винера. Это определяет выбор способа сканирования и соответствующей развертки оптимальных с точки зрения его реализационной сложности и сохранения фрактальных и статистических свойств изображения при его преобразовании к одномерному виду.An important advantage of the presented method is that in the image according to MNC (MNP), on the basis of the initial cell of the square-image (3 × 3 = 9) of the image with p = 1, formed according to a given criterion of relevance (i.e., according to information value for observer) in the image plane are performed in the form of topologically optimal controlled embeddings of the square-image facet pFas with respect to the parameter p = 1 ... 10 according to MNC. Moreover, the "grinding" of the image at the scanning stage is carried out to p = 0 (0Fas), i.e. to the pixel for which DBS is the statistics of the scanned series, immersed in the phase space of the image, is asymptotically reduced to a Gaussian distribution with zero mean and the corresponding standard deviation

which determines the possibility of using the Wiener filter when filtering the scanned series. This determines the choice of the scanning method and the corresponding scan that are optimal from the point of view of its implementation complexity and preservation of the fractal and statistical properties of the image when it is converted to one-dimensional form.

Claims (18)

1. A method for constructing and processing images on a plurality of colors in the Bayer arrangement for an optical imaging system using an image sensor having a plurality of elements of the optical system in the form of pixels to form a raw degraded image of an object (RAW), with each pixel of the image sensor for each portion of the image data of the corresponding set of colors in the Bayer arrangement is covered by one of a variety of color filters, including: a separation step in which the image data of a plurality of colors are divided by a plurality of color filters into a plurality of images of corresponding —R — G — B colors; - an image processing step in which reconstruction with a filter is sequentially performed for each part of the image data of the corresponding plurality of colors in the Bayer arrangement, separated in the separation step, using a plurality of respective image processing means, and then image processing is corrected, characterized in that, in addition contains - the stage of the minimum N-shaped scanning (MNC) of the RAW image - R - G - B colors, passed through the optical imaging system; the step of processing the colors of each pixel for the image data subjected to the reconstruction processing is performed by using the image data of a plurality of colors as if it were image data of the same color; - the stage of minimum N-shaped recursive scanning (MNP) of the processed image data; - the stage of restoration of each pixel of colors for the image in the Bayer arrangement; - at the initial stage of the formation of the minimum N-shaped scan (MNC), the unit square-image (K ₀ 1) with side 1 of the processed image is divided into nine congruent sub-squares and the corresponding unit segment [0; 1] (O ₀ 1) is also divided into nine equal segments and sequentially number the sub-squares (K ₁ 1; K ₁ 2; K ₁ 3; K ₁ 4, ..., K ₁ 9), K ₁ 1 neighbor K ₁ 2, K ₁ 2 neighbor K ₁ 3 and K ₁ 3 neighbor K ₁ 4, etc. and a segment into 9 identical parts (segments O ₁ 1, O ₁ 2, O ₁ 3, O ₁ 4, ..., O ₁ 9), the beginning of the curve is selected in the lower left corner, and the end is in the upper right and determines which segment in what sub-square falls into (while neighboring segments pass into sub-squares adjacent on the side), for each small segment, the coordinates of the points of the sub-square are determined, where its beginning, end and middle fall, then each small segment is divided into nine parts and similar constructions are carried out, etc. etc. until the size of congruent sub-squares is approximately equal to the size of a pixel, congruent sub-squares are determined equal to a pixel when forming the minimum initial cell of a unit square-image, scanning at the zero step of the initial cell (recursion direction) starts along the minimum N-shaped diagonal Peano curve with the 9th fractal genus , from the beginning in the lower left corner of the unit square, the beginning of the segment, to the upper right corner of the unit square-image, the end of the segment, at the first step, along the path with the bypass direction upward from the beginning of the unit square-image and further up, then turning to the right, down , down, right, up, up to the end of the unit square-image; the formed initial cell (3 × 3 = 9) at p = 1 is a facet (pFas), where p is the recursion step; - the minimum N-shaped scanning (MNC) of the RAW image of the object is carried out cyclically along a discrete square facet, built on the basis of scanning according to the rules of the curve filling the plane (SFC), with the setting of the beginning and direction of recursion, in the form of the minimum N-shaped diagonal Peano curve with The 9th fractal genus, for which the initial scanning cell (square-image) is set by a discrete initial unit square-image consisting of nine cells (3 × 3 = 9) and having four faces (sides), the beginning and end of which are located in opposite the vertices of the unit square-image; - bypassing sub-squares with minimal N-shaped scanning (MNC) is performed by transitions through the rotation operators (w) from one to the other according to the principle of proximity along the side, and the transition through a sub-square is carried out through its center, then a one-to-one correspondence between the segments and sub-squares is established, so that adjacent segments pass through the rotation operators (w) into adjacent (on the side) sub-squares, i.e. O ₁ 1 in K ₁ 1, O ₁ 2 in K ₁ 2, etc., to construct further directions of recursions, three types of rotation operators (w) for traversing are distinguished: the operator of rotation w1 up as the initial one (1Fas1) from the beginning of the unit square -image, the rotation operator w2 as a mirror from 1Fas1 to the right (1Fas2), the rotation operator w3 as a mirror from 1Fas2 down (1Fas3), and to obtain the subsequent direction of recursions, 2Fas is used (p = 2, with a side of 9 cells, 9 × 9 = 81), where the start specifies the end of 1Fas, etc .; - on the basis of the above operators of rotations w, they traverse in a sequence with the initial movement to a sub-square up from 1Fas and further: start → w1 → w1 → w2 → w3 → w3 → w2 → w1 → w1 → end, and the formation of the scanning curve is carried out by connecting the centers sub-squares with a line in the order of traversal, obtained on the second division, entering the sub-square K ₁ 1 in the lower left corner, and ending in the upper right corner, a Peano curve is built by sequentially traversing all the sub-squares obtained in the previous division, which remains the same (K ₁ 1 → K ₁ 2 → K ₁ 3 → K ₁ 4 →… → K ₁ 9) and redefines the location of the images of the points of the segment and the traversal rule in each sub-square K _n i up to a rotation similar to the traversal rule in the original unit square K ₀ 1, defining the Peano curve as a fractal; - at the processing stage, a wavelet transform is carried out in the form of a Haar transform over the signal value of each pixel from the scanned set of image states of the image sensor, the value of each pixel obtained by the Haar transform is fed to image processing for reconstruction using a filter in the form of a filter for solving the inverse problem from the optical the transfer function of the element of the optical system for forming RAW images, after which the image processing is corrected, and upon completion of the image processing, its inverse Haar transformation is performed over the plurality of images; - at the stage of the minimum N-shaped recursive sweep (MNP), the value obtained recovered after the inverse wavelet transform is stored in the memory block of the calculator F ^p according to the measurement coordinate of the storage number

consistent with scanning according to the MNP law, representing the plane in rotation coordinates (w), the carrier of which is the facet square pFas with the point of this plane

, having its center (xy), with a side of 3 ^p and an area of 9 ^p , where p = 0 ... N, while N sets the side of the unit square in which the size of the original image is inserted; at p = 0 (0Fas) the point degenerates into a pixel; for p> 0, the plane is divided into

independent points (sub-squares); parameters x, y for Q is its center on the plane; as a result, a multi-coordinate space of the point Q is formed: two coordinates of the Cartesian dimension (x, y), the coordinate of the rotation measurement (w), the coordinate of the dimension of the facet facet pFas, the coordinate of the measurement of the storage number

in the memory of the calculator F ^p ; common to these coordinates is the mapping of the Cartesian plane into memory with a one-to-one and continuous mapping of the point

into the memory of the calculator F ^p , and the transition to the next facet of the RAW image scanning occurs after the corresponding facet of the block scan in the memory of the calculator F ^{p is} completely filled, and the end of image restoration is carried out after filling the entire scan area of the image. 2. An image construction and processing system that implements the proposed method, which comprises an imaging unit using an image sensor having a plurality of pixels, wherein each pixel of the image sensor is covered with one of a plurality of color filters forming a separation unit for separating the image data of a plurality of colors of color filters to the image data with analog memory devices at the output corresponding to the colors of the color filters in the Bayer arrangement, the signal processing channel containing the signal processing unit, which is implemented on the corresponding filter obtained on the basis of the inverse function of the convolution of the filtered signal with the kernel in the form of the optical transfer function of the optical element of the imaging system, the signal correction unit, the input of which is connected to the output of the signal processing unit, and the output to the output of the signal processing channel, and the display unit, the outputs of the color separation unit are connected to the inputs image scanning unit, the information output of which is connected to the information input of the signal processing channel, characterized in that the image scanning unit consists of multiplexer units and an address setting unit for scanning a RAW image along the diagonal minimum N-shaped Peano curve, while maintaining the original topological and correlation the connectivity of the two-dimensional image, the input of the start of scanning is additionally introduced into the system, and a wavelet transform block is additionally introduced into the signal processing channel, for example, the Harr transform block, as well as an inverse Haar transform block, a scanning block containing a block of analog demultiplexers and a block of analog memory devices, and a display unit, in addition, ten signal delay elements and an "OR" element, and the address inputs of the multiplexer unit are connected to the outputs of the address setting unit, the information output of the image scanning unit is connected to the information input of the processing channel and a signal, which consists of a series-located Harr transformation unit, a signal processing unit, the information output of which is connected to the information input of the signal correction unit, the output of which through the signal processing channel output is connected to the information input of the Haar inverse transformation unit, the information output of the latter is connected to the input of the unit sweep, the input of which is connected to the information input of the demultiplexer unit, the address inputs of which are connected to the information outputs of the block for setting the addresses of the image scanning unit, and the information outputs of the demultiplexer unit are connected to the information inputs of the analog memory unit and then through the outputs of the scanner with the inputs of the display unit, the input the first signal delay element is connected to the recording inputs of the output analogue storage devices (ADS) of the color filters of the color separation unit, and the output of the first signal delay element is connected to the input of the installation in and a similar state of the address setting unit, the counting input of which through the second signal delay element with the timing input of the multiplexer unit of the image scanning unit, the output of the multiplexing end of the image scanning unit is connected to the timing input of the Haar conversion unit and through the third signal delay element to the timing input of the signal processing unit and the input of the fourth signal delay element, the output of which is connected to the clock input of the signal correction unit and through the fifth signal delay element to the inverse Haar conversion clock input of the inverse Haar conversion unit and then through the sixth delay element is connected to the clock input of the demultiplexer unit, the output of the sixth delay element is also connected through the seventh delay element with the clock input of the analog memory block of the scanner and then through the tenth signal delay element with the counting input of the block for setting the addresses of the image scanning unit i, and through the eighth signal delay element with the clock input of the display unit and then through the ninth signal delay element with the first input of the "OR" element, the second input of which is connected to the input of the system scan start, and the output to the input of the first signal delay element.

Images