DD298982A5 - METHOD AND CIRCUIT ARRANGEMENT FOR BETTER ADAPTIVE NOISE REDUCTION IN PICTURE PROCESSING - Google Patents

Abstract

The invention relates to a method and a circuit arrangement for adaptive noise reduction in image processing of nonstationary images. The aim of the invention is to increase the detection probability and recognizability of useful signals in distorted images by reducing noise. The object of the invention is to develop a method with circuit arrangement by means of which the noise in individual images is reduced and the result image is available without back transformation and further steps for glazing. According to the invention, the object is achieved by a method in which coefficients from overlapping subregions are determined from an image of any size from the signal amplitudes by educational regulations. In this case, the coefficient formation takes place in a method step according to the formation specification (x, y) 1 4 & pn (x, y) pn (xv, y)

Description

Field of application of the invention

The invention relates to a method and a circuit arrangement for the adaptation of noise detection in image processing of non-stationary images, bolsplolsweiso underwater Viclooblldor.

Characteristics of the known technical soundings

To mitigate vo.i disturbances In pictures a Violzahl of procedures and circuit -unordnungon are known, which were realized with different technical means.

So are t. Example, methods for noise reduction by means of zwoldimensional stationary filter or variable filter, which adapt to the spectral characteristics of the applied signal known.

According to the methods with stationary filters, e.g. operate on the principle of the Wiener filter, is that the noise reduction is optimal only for einolne image portions and strongo artefacts occur in other parts of the image, which show up by sanding the edges or poor noise suppression.

In the article Richter, Q., "For the Evaluation of Astronomical Imaging with the Automatic Flächonphotometer", Astronomical News, 299 (1978), p.283-303, a noise-mimicking ancestor using a 2 DH transformation is described, which is an adaptive filter, the meets the requirements of automatic detection of the signal characteristics with variable filters.

A disadvantage of this method is that before the image is displayed, it is absolutely necessary to pregone a back transformation out of the transformation space, whereby the desired characteristic of the filter, depending on the signal content, is averaged over different sized areas in the image, thus avoiding visual observation on the display Noise reduction device, a further step for smoothing is necessary and also by the block-by-block processing, the image format to be processed is bound to a multiple of the block size.

Object of the invention

Ziol the invention is to eliminate these disadvantages, to reduce the noise and to increase the detection probability and recognizability of useful signals in disturbed BiIcWn, especially underwater shots.

Daring the essence of the invention

The technical causes of the defects lie in the application of blockwise transformation methods and operations in the transformation space as well as the necessary inverse transformation.

The object of the invention is to develop a method and a circuit arrangement, by means of which an image of any format can be processed, the noise in individual images with nonstationary information content is reduced and the resulting image is without back transformation and without further steps * smoothing available.

According to the invention, the object is achieved by a method in which coefficients from overlapping subregions are generated from an image of any extent from the signal amplitudes.

The coefficients are determined by an education protocol that corresponds to a local operator and is called HOP 2 (H operator 2).

The HOP 2 consists of a procedural step with the educational requirement:

? h (x, y) = j [± p ⁿ (x, y) ± p ⁿ (x - v, y) ± (p ⁿ (x, yv) p ± ⁿ (x - v, y v) |.

For this, i is the direction of the coefficients in the considered image section and defines the corresponding signs of the expression. The distance v, the points used to form the step of the HOP 2 is dependent on the order η with ν = 2 ^{n and} 0: ä η ü N.

For the order η = 0, p ° (x, y) = ha (x, y).

According to the invention, an intermediate step is inserted before the processing of the HOP 2 in the order η = 0, which serves to symmetrize the environment of the observed pixel.

In this intermediate step, an intermediate image consisting of the mean values of adjacent points p 0 (x, y), p 0 (x + 1, y), p 0 (x, y + 1) and p 0 (x + 1, y + 1) is generated as follows:

h; (x, y) = 4-t ^{p0 (x'y )} + P ⁰ Ix + 1, y) + p ° ix, y + 1) + P ° (x + i, y + D]

According to the invention, higher-order input intermediate images have been generated for the spectral filtering from the input image, which are processed after the aforementioned method step. The input intermediate images are formed by averaging from the input intermediate image of the preceding order, ie in the formation specification all the signs are positive. The points p ⁿ (x, y), which are used for averaging, corresponding to η> 0, the mean values h "~ '(x, y), ie the mean values of the previous order η = η - 1.

According to the invention, the coefficients h for which unequal signs were defined according to the formation instruction are subjected to a subsequent test for significance.

According to the invention, there is significant signal structure in the test when at least one coefficient h? of the associated scanning in the test is significant.

According to the invention, the results are the most significant in the various orders n (OsnsN) in the form of a Mask superimposed on the input intermediate image such that in the output image only the significant Bildelemento the Input intermediate images of the respective order have been adopted. In accordance with the present invention, after dom leuton processing step of the last order to be processed which is not yet significant

set pixels of the output intermediate image with the image moments of the input intermediate image so that a complete output image is produced.

According to the invention, a circuit arrangement is also provided for the adaptive noise reduction of images and for the implementation of dos Method, at what or between dom Signalausgpng a Elngangszwlschenbildspelchers and dom signal input of a Auedahlgatters an operation gate with upstream Symmotrlorgatter and Eingangszwischonblldspolchor arranged

is what sbtastot dio signals of the input signal interleaver and associated center word and

DCF! Intermediate pictures are created, the data are stored in coefficient buffer memories and dio with a toostgattor

whose output signals are stored in a significance mask cache and fixed

Selection criteria in a Auswahlgatztor significant Signalolomento the input Zwischenbildspoichers in the Output intermediate image port has been transferred. According to the invention, the buffers for the mean values, significance and the output intermediate picture are over

interconnected feedback links are connected to the gates presented.

Ausführungsbeisplelo

At zwji examples, the inventions will be explained in more detail. Both examples assume image recordings of non-stationary images in the underwater area.

In the first example coefficients are formed in the ancestor aua don signal amplitudes of an image of any size. For the order η = 0, an intermediate image is generated from the input image, which consists of don average values of neighboring points p (x, y). p (x + 1, y), p (x, y + 1), p (x + 1, y + 1) consists of:

h; <x, y) - - ~ [p (x, y) + p (x + i, y) + p (x, y + D + p (x + i.y + D]

This leads to a symmetrization of the environment of the pixels considered. According to the following education regulation dos HOP 2

hS (x, y) = - ^ - (p ⁿ (x, y) + p ⁿ (x - v, y) + P "(x, y + v) + p ⁿ (x - v, y - v) ]

hl (x, y) = - (p ⁿ (x, y) - p ⁿ (x - v, y) + p ⁿ (x, y - v) - p ⁿ (x - v, y - v)] hj (x, y) - 4 "(P" ^(x 'v' ⁺ P " ^lx - v. y) - p ⁿ (x, y - v) - p ⁿ (x - v, y - v)]

h2 (x, y) = £ - [p ⁿ (x, y) - P ⁿ (x - v y) - p ⁿ (x, y - v) + p ⁿ (x - v, y - v)]

with ν = 2 ⁿ , where η dio is in order with OS η s N, averages hj (x, y) of symmetrical to the point p ⁿ (x, γ) are obtained

Environment, dio coefficients hj (x, y) as gradients in the x direction of the symmetrical to the point p "(x, y) lying environment, the Coefficients hj (x, y) as gradients in the y-direction of the symmetric to the point p ⁿ (x, y) lying environment and dio Coefficients hjlx, y) as curvature terms of the symmetry to the point p "(x, y) lying environment. For η = 0, p ° (x, y) = hö (x, y). The coefficients h £ (x, y), hj (x, y) and H2 (x, y) are subsequently subjected to a significance test. If at least one of the coefficients becomes significant in the test result, the value of the pixel p "(x, y) is significant, that is, it

there is a significant signal structure.

The result of the significance test is superimposed in the form of a mask on the input image such that only the significant Pixels have been adopted in an output intermediate image. For spectral filtering, the mean values h £ {x, y) for all coordinate points (x, y) of the image are set equal to p "* ¹ Ix, y) and

in the next order, the mean values hj * Mx, y) as well as the coefficients hj * ¹ Ix, y), h "* ¹ Ix,

y) and hj * '(x, y) holds.

The determined coefficients are subjected to a subsequent significance test, the result of which is the significance mask

modified.

The significant pixels p "* '(x, y) in this order, for which there was no significance in previous orders,

are adopted in the output intermediate. These process steps are repeated N times.

After the processing of the last method step of order N, all in the output intermediate picture do not yet take

significant pixels the mean values hjj (x, y).

The second example shows a possible circuit arrangement. According to the block diagram Fig. 1, this consists Circuit arrangement of three types of basic building blocks:

1. From two-dimensional memories 1 to 10 with any Abor within a system of the same format.

For digital signals, these memory z. B. can be realized as RAMs, with analog signals possibly by CCD arrays or by parallel galvanic connections.

2. From gates 14 to 19, which form one or more input images in one or more input images.

The input gate 14 is a changeover switch which switches the information from the input frame memory 1 at the order = 0 and the information from the feedback element 11 to the input intermediate frame memory 2 at η Φ Ο.

The balancing gate 15 consists of arithmetic units which, on the order N »0, realize the averaging from the values of the input picture field 2 and pass these averages on the intermediate intermediate picture pole 3.

For η Φ 0, the symmetry gate 1B directly switches the information from memory 2 to memory 3.

The HOP 2 gate 16 implements the HOP 2 education specification by selecting the pixels required for the HOP 2 education specification and correspondingly linking the samples. The elements of the gate 1 θ are arithmetic units which perform addition, substitution and division by four.

The test gate 17 consists of a discriminator (eg threshold switch or Fonsterdlskrlminator), which compare the output signals of the coefficient latches 6 to 7 with predetermined test sizes, wherein the logical Ausgangsslgnalo logically linked in an OR circuit and with dom word from feedback element 12 via a logical AND circuit, the output signal of the test gate 17 is generated.

The selection gate 18 consists of a switch which switches the information of the Zwangszwischonbildspelchers 2 on the Ausgangszwlschenb Idspelcher 9.

The control of this) switch is effected by the information from the significance mask buffer 8 and the information from eggs Rückführgllod 13th

After the last order to be processed η = N, the information of the input intermediate image memory 2 is directly switched to the output intermediate memory 9 if they have been allocated in previous orders.

Oae output gate 19 switches the output intermediate image of the output intermediate image memory 9 after processing the

last order η = N to the output frame memory 10. Otherwise, the output gate 19 is disabled. 3. From feedback elements 11 to 13, which place an image for the next order to be processed on the input of a logically preceding gate.

In principle, these feedback elements are sample and hold circuits and are used to realize the η = 0, ..., N spectral orders.

The feedback element 11 stores the values of the intermediate coefficient buffer 4, which are stored for the next successive order η (n> 0) via the input gate 14 as an input image in the input intermediate image memory 2.

The feedback element 12 stores the values of the significance mask intermediate 8 in order to provide them to the test gate 17 for the next orders η " η + 1.

The feedback element 13 stores the unused memory locations of the output intermediate image memory 9 and provides them for the next order η = η + 1 to the selection gate 18 for control.

For image processing, an input image which is present in the input image memory 1 is switched via the input gate 14 to an input intermediate image memory 2 and further via a balancing gate 15 to an input intermediate image memory 3. For order η => 0 symmetrization is realized in the balancing gate 15. For η> 0, this gate switches the input intermediate image 2 unchanged to the input intermediate-image memory 3. From there, the information is switched via the HOP-2 gate 16, which realizes the H-operator 2 (HOP 2) according to the educational specification and generates signals which are latched according to their characteristics to the coefficient latches 4 to 7. The contents of the coefficient buffers 5 to 7 are subjected to a signal check test in the test grid 17.

The result of the significance test is stored in a significance mask intermediate pointer 8. This significance mask controls the select gate 18 whose output signals are supplied to the output buffer memory 9. The output gate 19 switches after complete processing, i. at η = N, the output intermediate image from the output intermediate image memory 9 onto the output image memory 10 and thus provides the noise-reduced output image.

Claims (5)

1. A method for adaptive noise reduction in image processing of non-stationary images by means of temporally successive shots, in which the signal amplitudes of an image coefficients are formed, characterized in that are generated from an image of arbitrary extent from the signal amplitudes coefficients from overlapping sub-areas, in which so-called local operator (H-operator 2) from the delta points p (x, γ) a coefficient formation according to the education regulation h '(x, y) = -J- [± P ⁿ (x, y) + P ⁿ <x - v, y) ± p ⁿ (x, y - v) ± p ⁿ (x - v, y - v)] he follows. Here, i is the direction of the coefficients in the considered image section and defines the corresponding signs of the expression. The distance v, the points used to form the step of the HOP 2 is dependent on the order η with ν = 2 "and 0 s η s N, where for the order η = 0 p ° (x, y) = hj (x , y) and before the processing of the HOP 2 in the order η = 0 an intermediate step is inserted, which serves to symmetrize the environment of the pixel under consideration In this intermediate step an intermediate image, consisting of the mean values of adjacent points p ° (x, y), p ° (x + 1, y), p ° (x, y + 1) and p ° (x + 1, y + 1) are generated as follows: h>, y) = -J- [p ° (x, y) + Po (x + Ly) + p ° (x, v + D + p ° (x + i.y + D] 2. The method according to item 1, characterized in that are generated and processed for a spectral filtering from an input image input intermediate images of higher order, which arise by averaging according to the education rule, all signs are positive and the averaging used points p ⁿ (x, y) at η> O correspond to the mean values of the previous order η = η - 1. 3. Method according to item 1, characterized in that the coefficients h ", for which unequal signs were defined, are subjected to a subsequent test for significance, in which significant signal structure exists, if at least one coefficient h" of the associated sample value is significant in the test and the result of the significance test in the various orders η (O ^ η <N) is superimposed in the form of a mask on the input intermediate image such that only significant picture elements of the input intermediate images of the respective order are taken over into the output image and after the last processing step the last order to be processed (n = N), the not significantly set pixels of the output intermediate image are occupied by the pixels of the input intermediate image, so that a complete output image is formed. 4., Adaptive noise reduction circuit arrangement in image processing and for carrying out the method according to item 1, characterized in that between the signal output of an input intermediate image memory (2) and the signal input of a selection gate (18) an operator gate (16) is arranged, which the signals of Input intermediate image memory (2) via an upstream balancing gate (15) and a second input intermediate image memory (3) and generates associated average and coefficient intermediate images, which are stored in coefficient buffers (5; 6; 7) and connected to a test gate (17) whose Output signals are stored in a significance mask latch (8) and predetermined signal elements of the input buffer memory (2) are transferred under control of the significance mask buffer (8) to the output buffer memory (9), the latches (4; 8, 9) via intermediate feedback members (11; 12; 13) are connected to the upstream gates (14; 17; 18). 5. The circuit arrangement according to item 4, characterized in that at η Φ 0, the balancing gate (15), the information from the input intermediate image memory (2) to the second input intermediate image memory (3) directly switches. For this 1 page drawing