KR101331052B1 - Device for improving stereo matching results, method for improving stereo matching results using the device and system for receving stereo matching results - Google Patents

Abstract

An apparatus for improving stereo matching results is disclosed. The apparatus for improving the stereo matching result includes a stereo camera unit for outputting binocular disparity images using binocular disparity between two images preprocessed according to a plurality of preprocessing conditions, and a discrete cosine transform on the binocular disparity images. Stripe evaluation for inputting a discrete cosine transform unit for generating DCT coefficients by DCT) and the generated DCT coefficients and evaluating the amount of stripes distributed on the screen using AC coefficients including the stripe pattern among the DCT coefficients. A predetermined part of the DCT coefficients obtained by the condition evaluator and the discrete cosine transform unit for evaluating an optimal condition for a pretreatment condition corresponding to the lowest stripe amount among the evaluated stripe amounts among the plurality of pretreatment conditions. Binocular with the streaks removed by changing AC coefficients and transforming the altered DCT coefficients by inverse discrete cosine Generating a difference image comprises removing parts of stripes.

Description

DEVICE FOR IMPROVING STEREO MATCHING RESULTS, METHOD FOR IMPROVING STEREO MATCHING RESULTS USING THE DEVICE AND SYSTEM FOR RECEVING STEREO MATCHING RESULTS }

The present invention relates to an apparatus for improving a stereo matching result and a method using the apparatus, and more particularly, to an apparatus for improving the quality of a binocular disparity image that is the stereo matching result and a method using the apparatus.

The present invention is derived from a study conducted as part of the IT source technology development project of the Ministry of Knowledge Economy [Task management number: 2008-F-037-01, Task name: u-robot HRI solution and core device technology development].

There are several ways to calculate binocular disparity between two images as a result of stereo matching. Existing binocular disparity can be calculated by using ASIC, Dynamic Programming Stereo (DPS) and Scanline Smoothness Optimization.

However, these methods have a disadvantage in that streaks noise occurs in the binocular parallax image due to an error occurring in the unit of the scanning line of the image, and the noise degree varies according to the alignment condition of the binocular camera.

FIG. 1 is a diagram illustrating a binocular parallax image in which a conventional stripe is generated, and FIG. 2 is a view illustrating an 8 × 8 block in which many stripes are present in FIG. 1.

As shown in FIG. 1, streaks generated in the binocular parallax image obtained by optimizing one-dimensional energy are formed in a horizontal direction, and as shown in FIG. There is a characteristic.

3 is a diagram illustrating an image of a result of Fourier transforming the binocular parallax image illustrated in FIG. 1.

Referring to FIG. 3, when the image including the horizontal stripes is Fourier-transformed and the converted result image is observed on a log scale, vertical components appear strongly, but as shown in FIGS. 1 and 2, respectively. Since the length of is not long, a lot of high frequency components in the horizontal direction also appear.

Conventional techniques for removing streaks in such images detect streaks that appear as straight lines in the Fourier region and remove the detected streaks using linear regression analysis.

However, this conventional technique is limited to removing streaks across the original image. Because, as shown in FIG. 4, streaks that are locally short in the image contain a large number of frequency components along the direction of travel of the streaks, so as shown in FIG. 5, they occupy a very wide range in the Fourier region. do. Therefore, linear regression analysis, which is used in existing techniques, cannot remove streaks that exist locally in the image.

Other existing techniques for reducing streaks across the screen require computation of many pixels in the image, which not only slows down processing but also eliminates local short streaks in the image.

Accordingly, an object of the present invention is to provide an apparatus for improving stereo matching results by reducing horizontal streaks present in binocular parallax images.

Another object of the present invention is to provide a method for improving stereo matching results using the apparatus.

Another object of the present invention is to provide a system for receiving the stereo matching result.

In order to achieve the above object, an apparatus for improving stereo matching results according to an aspect of the present invention includes a stereo camera unit for outputting a binocular parallax image between two images preprocessed according to a plurality of preprocessing conditions, and the binocular parallax image. Discrete Cosine Transform (DCT) to generate a DCT coefficient by using a discrete cosine transform (DCT), and using the generated DCT coefficients, AC coefficients containing a stripe pattern component in a predetermined direction of the DCT coefficients A stripe evaluator for evaluating the amount of stripes distributed on the screen, a condition evaluator for evaluating a pretreatment condition corresponding to the lowest stripe amount among the evaluated stripe amounts among the plurality of pretreatment conditions as an optimal condition, and the discrete cosine transform Change predetermined AC coefficients among the DCT coefficients obtained by the negative and inverse discrete cosa And a stripe removing unit to generate a binocular parallax image from which the stripe is removed according to the optimal condition.

According to another aspect of the present invention, there is provided a method of improving a stereo matching result, including: outputting a binocular parallax image between two images preprocessed according to a plurality of preprocessing conditions, and performing a discrete cosine transform on the binocular parallax image: Generating DCT coefficients, evaluating the amount of stripes distributed on the screen by using AC coefficients corresponding to the stripe pattern among the generated DCT coefficients, and the estimated amount of stripes among the plurality of preprocessing conditions. Evaluating the pretreatment condition corresponding to the lowest stripe amount among the optimal conditions, changing the AC coefficients corresponding to the stripe pattern, and inverse discrete cosine transforming the changed DCT coefficients according to the optimal condition. Generating the binocular parallax image from which streaks have been removed.

According to another aspect of the present invention, a reception system for receiving a stereo matching result receives a binocular parallax image including an original image and a stripe pattern as a compression-encoded bitstream, decodes the bitstream to restore a DCT coefficient, A compressed image decoder for reconstructing the binocular disparity image from which the stripe pattern is reduced by changing AC coefficients of a block corresponding to the binocular disparity image among the reconstructed DCT coefficients, and performing inverse discrete cosine conversion of the changed DCT coefficients; And a de-blocking filter for de-blocking a block included in the parallax image.

According to the present invention, in a stereo matching system in which horizontal streak noise occurs, the amount of streaks can be relatively evaluated, and a condition in which the amount is minimized can be easily found, and a binocular parallax image with reduced horizontal streaks can be obtained according to this condition. have.

It can also take advantage of discrete cosine transformers (DCTs) in software versions used in image processing applications for still or video processing, or discrete cosine transformers included in video hardware codecs, so that even low-performance embedded boards can be processed. Is possible.

1 is a diagram illustrating a binocular parallax image in which a conventional stripe is generated.
FIG. 2 is a diagram illustrating a region in which a lot of stripes occur in FIG. 1 in units of 8 × 8 blocks.
FIG. 3 is a diagram illustrating a result image of applying a Fourier transform to the binocular parallax image illustrated in FIG. 1.
4 is a diagram illustrating an image including short horizontal stripes.
FIG. 5 is a diagram illustrating a Fourier transform result of the short horizontal stripes illustrated in FIG. 4.
6 is a block diagram illustrating an apparatus for improving a stereo matching result according to an embodiment of the present invention.
FIG. 7 is a diagram illustrating a binocular parallax image output from the stereo matching unit illustrated in FIG. 6.
8 shows a basic pattern of DCT of 8 × 8.
FIG. 9 is a graph showing output values of the stripe evaluator shown in FIG. 6 for three binocular parallax images having different stripe distributions according to different preprocessing conditions.
FIG. 10 is a diagram illustrating a binocular parallax image in which stripes are reduced by the stripe removing unit illustrated in FIG. 6.
11 and 12 are flowcharts illustrating a method of improving a stereo matching result according to an embodiment of the present invention.
13 is a block diagram illustrating an apparatus for improving a stereo matching result according to another embodiment of the present invention.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

6 is a block diagram illustrating an apparatus for improving a stereo matching result according to an embodiment of the present invention.

Referring to FIG. 6, the apparatus 100 for improving a stereo matching result according to an embodiment of the present invention includes a stereo camera unit 110, a discrete cosine transformer (DCT) 120, and stripes. ) Includes an evaluator 130, a condition evaluator 140, and a stripe remover 150.

The stereo camera unit 110 generates a binocular disparity image by calculating a binocular disparity between two images. In order to generate the binocular parallax image, the stereo camera unit 110 includes a camera unit 112, a preprocessor 114, and a stereo matching unit 116. The camera unit 112 includes a right camera unit 112A and a left camera unit 112B whose optical axes are aligned in parallel with each other, and the right camera unit 112A generates a right image, and the left camera unit ( 112B generates a left image.

The preprocessor 114 receives a right image and receives a right image preprocessor 114A that performs a preprocessing process on the right image and the left image and performs a preprocessing process on the left image. Section 114B. The preprocessing process includes an independent brightness adjustment process, a histogram smoothing process, a parallel movement process, and an image rotation process of two images. Since these processes are performed before the stereo matching process, the stereo matching unit 116 is performed. The performance of stereo matching performed at may be improved.

The stereo matching unit 116 receives the preprocessed right and left images, calculates binocular disparity between the images using an algorithm such as a dynamic programming technique, and calculates the calculated binocular disparity between the brightness information and the brightness information. Generates and outputs a binocular parallax image represented by. In this case, the stereo matching unit 116 outputs binocular parallax images having different error distributions according to preprocessing conditions, and is a one-dimensional energy optimization technique capable of implementing hardware on the binocular parallax image having an error distribution. When the energy optimization technique is applied, the error distribution is mainly in the form of streaks. Hereinafter, a binocular parallax image having different stripe distributions according to preprocessing conditions will be described with reference to FIG. 7.

FIG. 7 is a diagram illustrating an example of a binocular parallax image having different stripe distributions output from the stereo matching unit illustrated in FIG. 6.

In FIG. 7, three binocular parallax images CASE 1, CASE 2, and CASE 3 having different stripe distributions are displayed according to different preprocessing conditions. The different pretreatment condition in this embodiment is a change in the vertical alignment position.

Specifically, CASE 1 moves the left image up one line so that the i + 1th horizontal line (left [i + 1]) from the top of the left image matches the ith horizontal line (right [i]) from the top of the right image. It is a binocular parallax image matched by parallel movement.

CASE 2 is a binocular parallax image matched using the left image as it is so that the i-th horizontal line (left [i]) from the top of the left image matches the i-th horizontal line (right [i]) from the top of the right image.

CASE 3 moves the left image down by one line so that the i-1th horizontal line of the left image (left [i-1]) matches the ith horizontal line of the right image (right [i]). Parallax image.

In FIG. 7, streaks appear most severely in the binocular parallax image of CASE 1, and streaks appear least in the binocular parallax image of CASE 3.

In this embodiment, the vertical alignment position is described as an example of the preprocessing condition. However, as another example of the pretreatment condition, the left camera's exposure value is changed with respect to the right camera's exposure value, and the left camera's luminance gain with respect to the luminance gain of the right camera is described. It may include various manipulations that affect the binocular parallax image, such as changing the luminance gain.

Referring to FIG. 6 again, the DCT 120 divides the binocular parallax image into block units having a predetermined size. Thereafter, a discrete cosine transform is performed on each of the divided blocks to output DCT coefficients of the respective blocks. For example, the DCT 120 divides the received binocular parallax image into k (natural numbers) 8 (vertical pixel) X 8 (horizontal pixel) block images, and each block image (X1 to Xk) is discrete cosine. By transforming, discrete cosine transform (DCT) coefficients Y1 to Yk are calculated.

The computed discrete cosine transform (DCT) coefficient Yk may be represented by Equation 1 below.

Here, i is a vertical pixel coordinate in the input binocular parallax image block Xk, and j is a coordinate of a horizontal pixel in the binocular parallax image block Xk. In addition, x and y represent the index of a DCT coefficient.

FIG. 8 is a diagram showing a basic pattern of DCT of 8 X 8, wherein Yk _{x, y in} Equation 1 _, wherein the subscript x represents x = 0, 1, ..., 7 in a downward direction from the top, The subscript y represents the component represented by y = 0, 1, ..., 7 from left to right.

As shown in FIG. 8, the DCT coefficients Yk _{x and y} of the block Xk may be used as a measure of how many noise pattern components are in the block Xk.

Discrete Cosine Transform (DCT) has the following advantages over Discrete Fourier Transform (DFT).

First, the discrete Fourier transform process (DFT) generates high frequency components by the discontinuity of the left and right and top and bottom boundaries of the image in the process of creating a periodic image, but the discrete cosine transform technique produces a periodic image. Since the image is symmetrical to the left and right or up and down in the process, discontinuity does not occur and high frequency components due to discontinuity do not occur. Therefore, by using the DCT, the high frequency component of the stripe noise can be detected more effectively than the conventional technique of finding the high frequency component corresponding to the stripe component in the binocular disparity image using the Fourier transform.

Second, DFT always generates magnitude and phase information, but DCT does not generate phase information. In detecting fringes of binocular parallax images, DCT is advantageous over DFT because phase information is not needed.

Third, unlike conventional techniques of detecting stripes, DFT divides an image into small blocks of 8 x 8 or 16 x 16, and performs discrete cosine transform on each block. It is only included within the block and has the advantage of not affecting adjacent blocks.

Referring back to FIG. 6, the stripe evaluator 130 receives the DCT coefficients of k blocks from the DCT 120 and evaluates the amount of stripes distributed on the screen using AC coefficients. The zeroth order coefficients (Yk _{x , y} = x = y = 0) of the DCT coefficients are called DC coefficients because they are not cosine functions unlike other coefficients, and the other coefficients except for the zeroth order coefficients are AC coefficients.

만이 사용될 수도 있다. 다시 말해, 상기 줄무늬 평가부(130)는 상기 양안 시차 영상을 상기 8 x 8 블록 크기로 분할한 경우, 줄무늬 패턴에 대응하는 주파수 성분이 가장 높은 좌측 맨 아래쪽 AC 계수들(도 8을 참조)을 이용하여 화면에 분포한 줄무늬 양을 평가한다. Of these AC coefficients, y = 0 is a good representation of the horizontal line. When discrete cosine transforms k 8 X 8 block images, it is the thinnest horizontal horizontal frequency component of all k blocks.

Only may be used. In other words, when the binocular parallax image is divided into the 8 × 8 block size, the stripe evaluator 130 obtains the lower left bottom AC coefficients having the highest frequency components corresponding to the stripe pattern (see FIG. 8). Evaluate the amount of stripes distributed on the screen.

The amount of stripes for the k blocks, that is, the stripe evaluation function f, may be expressed as a sum of absolute values, for example, as shown in Equation 2 below.

The stripe evaluation function f calculates a consistent evaluation result for the binocular parallax images CASE 1, CASE 2, and CASE 3 of FIG. 7 and the binocular parallax images of the next frame as follows.

FIG. 9 is a graph showing output values of the stripe evaluator shown in FIG. 6 for three binocular parallax images having different stripe distributions according to different preprocessing conditions.

FIG. 9 shows first to third graphs G1, G2, and G3, and each of the first to third graphs G1, G2, and G3 includes binocular parallax images of CASE 1, CASE 2, and CASE 3 of FIG. 7. It is a graph which shows the output value of the stripe evaluation part 130 with respect to the field. Accordingly, the first graph G1 is an output value of the stripe evaluator (130 of FIG. 6) for the binocular parallax image having the greatest stripe distribution, and the third graph G3 is a binocular parallax image having the lowest stripe distribution. The output value of the stripe evaluation part is shown.

Referring back to FIG. 6, the condition evaluator 140 stores output values output from the stripe evaluator 130 according to a preprocessing condition according to the vertical alignment position. The condition evaluator 140 compares the stored output values with each other and outputs the preprocessing condition in the case of outputting the lowest output value as an optimum condition. That is, in the case of FIG. 9, both conditions of the third graph G3, that is, the CASE 3 of FIG. 7, in which the condition evaluator 140 forms a distribution of output values that are consistently smaller than those of the other graphs G1 and G2. The preprocessing condition applied to the parallax image is determined as an optimal condition, and the binocular parallax image corresponding to the optimal condition, that is, the binocular parallax image of CASE 3 of FIG. 7, is evaluated as the optimal binocular parallax image.

According to the evaluation result of the condition evaluator 140, the stripe remover 150 of FIG. 6 performs discrete cosine transform on the optimal binocular disparity image generated by the stereo matching unit 116 by the DCT 120. The result value is input, and the AC coefficients corresponding to the predetermined condition are changed (or deleted) to 0 or a value close to 0 in the result value. Then, the inverse discrete cosine transforms the altered AC coefficients to reconstruct the image.

FIG. 10 is a diagram illustrating a binocular parallax image in which stripes are reduced by the stripe removing unit illustrated in FIG. 6.

Referring to FIG. 10, a binocular parallax image obtained by changing the AC coefficients from the DCT coefficients of the binocular parallax image of case 3 of FIG. 7 by the stripe removing unit 150 and performing inverse discrete cosine conversion of the changed AC coefficients is illustrated in FIG. It can be seen that the streak noise is significantly reduced compared to the binocular parallax image of CASE 3 of 7.

On the other hand, since the binocular parallax image obtained by the stripe remover 150 is a result calculated in units of blocks, the binocular parallax image obtained by the stripe remover 150 has a prominent block boundary. Therefore, the apparatus for improving the stereo matching result according to an embodiment of the present invention may further include a de-blocking filter (not shown) for performing a de-blocking process on the boundary of each block. .

11 and 12 are flowcharts illustrating a method of improving a stereo matching result according to an embodiment of the present invention. For better understanding of the description, reference is made to FIG. 6 together.

6 and 11, first, a preprocessing condition set in the stereo camera unit 110 is initialized in order to output a binocular parallax image (S1110).

Subsequently, the right image and the left image are acquired by the right camera 112A and the left camera 112B of the stereo camera unit 110, respectively (S1112).

Subsequently, the preprocessing process is performed on the obtained right and left images according to predetermined preprocessing conditions. A stereo matching process using binocular disparity between the preprocessed right image and the left image is performed, and a binocular disparity image, which is an execution result, is generated (S1114). At this time, the pretreatment conditions are variously manipulated. Various operations of these preprocessing conditions include changing the condition according to the vertical alignment position between the two images, changing the exposure value of the left camera to the exposure value of the right camera or the variable of the exposure value of the right camera to the exposure value of the left camera, and the luminance gain of the right camera ( manipulation affecting binocular parallax images such as the luminance gain of the left camera for luminance gain or the change in the luminance gain of the right camera for luminance gain of the left camera.

Subsequently, the generated binocular parallax image is divided into blocks having a predetermined size, and a discrete cosine transform process is performed on each of the divided blocks (S1116).

Subsequently, the stripe evaluation is performed on each block image having the discrete cosine transform (S1118), and the stripe evaluation result and the preprocessing condition are accumulated and stored (S1120).

Thereafter, if there is a remaining pretreatment condition, the current pretreatment condition is changed to another one of the remaining pretreatment conditions (S1124), and the processes (S1110, S1112, S1114, S1116, S1118 and S1120) according to the changed pretreatment condition. This is done again. If there is no remaining pretreatment condition, that is, the amount of stripes of the binocular parallax image to which all pretreatment conditions are applied is evaluated, and an optimal pretreatment condition is determined from the evaluated result (S1126).

Thereafter, if the data is not sufficient (S1128), the process returns to the first step (S1110), and if the data is sufficient (S1128), the evaluation result of the stripe evaluation according to the pretreatment condition is selected, and the optimal pretreatment condition is selected (S1130). According to the selected optimal preprocessing condition, the left image and the right image are generated in the camera module 112 of the stereo camera unit 110 of FIG. 6 (S1134).

Subsequently, a binocular disparity image according to the optimal preprocessing condition is generated using the generated binocular disparity between the left image and the right image (S1136), and the generated binocular disparity image is a block image of a predetermined size (8 × 8). In this case, a discrete cosine transform process is performed on the divided block images (S1138).

Subsequently, some AC coefficients are changed in the DCT coefficients of the discrete cosine transformed binocular parallax image by the stripe removing unit 150 of FIG. 6, and an inverse discrete cosine transform process is performed on the changed AC coefficients. Accordingly, a binocular parallax image with reduced streaked noise is generated (S1140).

Subsequently, since the binocular parallax image acquired by the stripe remover 150 is a result image calculated in units of blocks, a de-blocking process is performed on the boundary of each block (S1142).

13 is a block diagram illustrating an apparatus for improving a stereo matching result according to another embodiment of the present invention.

Referring to FIG. 13, an apparatus 200 for improving a stereo matching result according to another embodiment of the present invention includes a stereo camera unit 210, a compressed image encoder 220, and a compressed image decoder 230.

When the output result of the stereo camera system needs to be transmitted through a network having a limited bandwidth, such as a wireless LAN or a USB, it is preferable that a compressed image in which duplicate data is removed is transmitted in a network transmission interval. In this case, the DCT 120 in the embodiment of FIG. 6 corresponds to the compressed image encoder 220 including the DCT of FIG. 13, and the stripe remover 150 of FIG. 6 decodes the compressed image. It may correspond to 230.

In the embodiment of FIG. 6, the stripe evaluation unit 130 may be omitted in the embodiment of FIG. 13 because it is necessary only when the optimum condition is found in the design of the stereo camera unit 110.

The compressed image encoder 220 is, for example, a video hardware encoder of a standard such as MPEG2, MPEG4, or H.263, or a JPEG encoder, and is a frame in which a pre-processed right image or a pre-processed left image and a binocular parallax image including stripes are combined. Receive the video. The extruded image encoder 220 applies a discrete cosine transform to the received frame image in units of 8 × 8 blocks, then quantizes DCT coefficients, and variable length encoding of the quantized DCT coefficients. The bit stream is generated and output through the process.

The compressed video decoder 230 may variable length decode the decoded video bitstream received through the network, and dequantize the decoded video bitstream to generate a DCT coefficient. The stripe removing unit included in the compressed image decoder 230 changes AC coefficients corresponding to a predetermined condition among AC coefficients of blocks in the binocular parallax image region to zero or a value close to zero. Thereafter, the image is reconstructed by performing inverse DCT transformation on the changed AC coefficients. Thereafter, an image obtained by applying a de-blocking filter to the boundary of each block in the reconstructed image is output.

Using the embodiment of FIG. 13, streaks of binocular disparity images may be removed without compromising the quality of the original image encoded together.

Claims (16)

A stereo camera unit configured to output a binocular parallax image between two images preprocessed according to a plurality of preprocessing conditions;
A discrete cosine transform unit for generating a DCT coefficient by performing discrete cosine transform (DCT) on the binocular parallax image;
A stripe evaluation unit configured to receive the generated DCT coefficients and evaluate an amount of stripes distributed on a screen by using AC coefficients including stripe pattern components in a predetermined direction among the DCT coefficients;
A condition evaluator configured to evaluate a pretreatment condition corresponding to the lowest stripe amount among the evaluated stripe amounts among the plurality of pretreatment conditions as an optimal condition; And
Alternating AC (Alternating Current) coefficients corresponding to a predetermined condition among the DCT coefficients obtained by the discrete cosine transform unit, and performing inverse discrete cosine conversion on the changed DCT coefficients to remove the streaks according to the optimum condition. Stripe elimination unit for generating binocular parallax image
Apparatus for improving a stereo matching result comprising a.
The method of claim 1, wherein the stereo camera unit,
And outputting the binocular parallax image according to a preprocessing condition according to the vertical alignment position between the two images.
The method of claim 1, wherein the stereo camera unit,
The binocular parallax image is taken according to the preprocessing conditions including a condition for setting the exposure value of the other camera with respect to the exposure value of one camera and a condition for setting the luminance gain of the other camera with respect to the luminance gain of the one camera. And outputting a stereo matching result.
The method of claim 1, wherein the discrete cosine transform unit,
And dividing the binocular disparity image into blocks of a predetermined unit and performing the discrete cosine transform on each of the divided blocks.
The method of claim 4, wherein the discrete cosine transform unit,
Dividing the binocular parallax image into one block unit of 8 (number of vertical pixels) x 8 (number of horizontal pixels) and 16 (number of vertical pixels) x 16 (number of horizontal pixels) blocks. Characterized by a device for improving stereo matching results.
The method of claim 5, wherein the stripe evaluation unit,
The binocular parallax image In the case of dividing K (integers greater than zero) into the 8 x 8 blocks, the amount of stripes distributed on the screen is obtained by using each AC coefficient having the highest frequency component corresponding to the stripe pattern among the DCT coefficients of each 8 x 8 block. Evaluating a stereo matching result.
The method of claim 1, wherein the stripe removing unit,
After changing each of the AC coefficients corresponding to the predetermined condition to 0, the apparatus for improving the stereo matching result, characterized in that to generate the binocular parallax image from which the stripe is removed by inverse discrete cosine transform the DCT coefficients .
8. The stereoscopic apparatus of claim 7, further comprising a de-blocking filter for de-blocking a block boundary included in the binocular disparity image reconstructed in units of blocks by the inverse discrete cosine transform unit. Device for improving matching results.
2. The apparatus of claim 1, wherein the stripe pattern in a predetermined direction is a stripe pattern in a horizontal direction. 3.
Outputting a binocular parallax image between two images preprocessed according to a plurality of preprocessing conditions;
Generating a DCT coefficient by performing Discrete Cosine Transform (DCT) on the binocular parallax image;
Evaluating the amount of stripes distributed on the screen by using alternating current (AC) coefficients corresponding to the stripe pattern among the generated DCT coefficients;
Evaluating a pretreatment condition corresponding to the lowest stripe amount among the evaluated stripe amounts among the plurality of pretreatment conditions as an optimal condition; And
Changing the AC coefficients corresponding to the stripe pattern and inverse discrete cosine transforming the changed DCT coefficients to generate the binocular parallax image from which the stripe is removed according to the optimum condition
How to improve the stereo matching result comprising a.
The method of claim 10, wherein generating the DCT coefficients,
Dividing the binocular parallax image by a predetermined block unit; And
And performing the discrete cosine transform on each partitioned block.
The method of claim 11, wherein dividing the data into predetermined block units comprises:
Dividing the binocular parallax image into 8 (number of vertical pixels) x 8 (number of horizontal pixels) blocks.
The method of claim 12, wherein the step of evaluating the amount of stripes,
The binocular parallax image In the case of dividing K (integers greater than zero) into the 8 x 8 blocks, the amount of stripes distributed on the screen is obtained by using each AC coefficient having the highest frequency component corresponding to the stripe pattern among the DCT coefficients of each 8 x 8 block. Evaluating stereo matching results.
The method of claim 10, wherein the generating of the binocular parallax image comprises:
And changing each of the AC coefficients corresponding to a predetermined condition to 0, and then inverse discrete cosine transforming the DCT coefficients to generate the binocular parallax image from which the streaks have been removed.
Receiving a binocular parallax image including an original image and a stripe pattern as a compression-encoded bitstream, decoding the bitstream to reconstruct DCT coefficients, and AC of a block corresponding to the binocular parallax image among the reconstructed DCT coefficients; A compressed image decoder configured to change coefficients and inverse discrete cosine transform the changed DCT coefficients to reconstruct the binocular parallax image having the reduced stripe pattern; And
De-blocking filter for de-blocking the blocks included in the reconstructed binocular parallax image
Receiving system for receiving a stereo matching result comprising a.
The method of claim 15, wherein the compressed video decoder,
After changing each of the AC coefficients corresponding to a predetermined condition among the restored DCT coefficients to 0, the stereo matching result of restoring the binocular disparity image of which the stripe pattern is reduced by inverse discrete cosine transforming the DCT coefficients. Receiving system for receiving.

Images