KR20000032095A - Moving object division device and method of compression image - Google Patents

Abstract

PURPOSE: A moving object division device and method of compression image is provided to directly divide a moving object of a compression image without completely decoding the image compressed by MPEG-1/2 compression standard. CONSTITUTION: A moving object division device of compression image comprises a DC image reconstructor(1), a DC image divider(2), a moving vector extractor(3) and a time-space divider(4). The DC image reconstructor(1) decodes a portion of bit columns of a compression image and extracts DC coefficient of discrete cosine transform per respect block to reconstruct it as a compressed DC image. The DC image divider(2) divides a moving object using the reconstructed DC image. The moving vector extractor(3) extracts a moving vector by decoding a portion of bit columns of the compression image. The time-space divider(4) remove a mis-extracted moving object using the moving vector.

Description

Moving object segmentation device and method of compressed image

The present invention relates to a moving object segmentation apparatus and method for indexing surveillance images compressed by moving picture experts group-1,2 (MPEG-1,2), which is an image compression standard, on an object basis.

Conventional moving object segmentation techniques are performed on an uncompressed original image. To apply such a moving object segmentation technique to a compressed image, it is necessary to completely decode a bit string of a compressed image and segment a moving object using the decoded image. Therefore, in order to segment a moving object in order to object-based index the compressed image, a decoding time of the compressed image is required separately, and a large capacity memory is required because the data amount of the decoded image increases.

The conventional method for object-based indexing of an image compressed with MPEG-1 / 2 extracts a motion vector from a bit stream of a compressed image, constructs a motion vector field in units of frames, and then divides the motion vector field. However, if there is no motion vector in the bit stream of the compressed image, the moving object cannot be divided. In addition, if a noise vector is present in the background without motion, the noise vector is also recognized as a motion vector, and the performance of moving object segmentation is degraded. Therefore, the motion vector of the compressed image does not accurately represent the actual motion of the object, and even in the background of no motion, the segmentation performance is deteriorated due to the noise vector. Since the motion vector is provided in units of macroblocks, it is difficult to divide a small moving object. There was a problem.

Accordingly, the present invention has been devised to solve the above problems of the prior art, and does not completely decode the bit stream of an image compressed by the MPEG-1 / 2 compression standard, but stably decomposes a moving object of the direct compressed image. It is an object of the present invention to provide a moving object segmentation apparatus and method for segmenting compressed images.

1 is a block diagram showing a moving object segmentation apparatus of a compressed image according to an embodiment of the present invention;

2 is a flowchart illustrating a moving object segmentation method of a compressed image according to an embodiment of the present invention;

3 is a detailed flowchart of the space-time partitioning scheme of FIG. 2.

※ Explanation of code about main part of drawing ※

1: DC image reconstruction unit 2: DC image division unit

3: motion vector extractor 4: space-time partition

In order to achieve the above object, the present invention, a DC image reconstruction unit for reconstructing the compressed image of each frame unit into a DC image using the DC coefficient of the discrete cosine transform (Discrete Cosine Transform), and the reconstructed DC image A DC image splitter for dividing a motion object using the motion image extractor, a motion vector extractor for partially decoding the compressed image to extract a motion vector, and separating two adjacent motion objects by using the motion vector. It provides a moving object segmentation apparatus of a compressed image, characterized in that it comprises a space-time segmentation unit to remove.

The present invention also provides a first step of partially decoding a compressed image frame-by-frame, extracting a DC coefficient of a discrete cosine transform for each block, and reconstructing it into a DC image, and a second vector extracting a motion vector by partially decoding the compressed image. And a third step of dividing a motion object by using the reconstructed DC image, and a fourth step of separating two adjacent motion objects by using the motion vector and removing the wrongly extracted motion object. A moving object segmentation method of a compressed image is provided.

The present invention also provides a first step of partially decoding a compressed image on a frame-by-frame basis, extracting a DC coefficient of a discrete cosine transform for each block, and reconstructing it into a DC image, and extracting a motion vector by partially decoding the compressed image. A second step of setting a DC image of the first frame of the compressed image as a reference image, a fourth step of obtaining a difference image between the DC image of another frame and the reference image, and low pass filtering of the difference image A fifth step of reconstructing a binary image composed of a motion pixel and a still pixel by performing low pass filtering, a sixth step of grouping neighboring motion pixels by median filtering the binary image and dividing them into a moving object; A seventh step of dividing a motion object into two regions by using the motion vector; a magnitude of a motion vector and a flatness of a difference image for each motion object An eighth step of calculating the mean and the variance, and if the magnitude of the motion vector of each moving object is smaller than the first threshold value, the difference of the difference image is smaller than the second threshold value, and the variance of the difference image is smaller than the third threshold value, In addition, the present invention provides a computer-readable recording medium having recorded thereon a program for executing a ninth step of determining and removing a moving object as an unnecessary object and improving partitioning performance.

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

1 is a block diagram illustrating a moving object segmentation apparatus of a compressed image according to an embodiment of the present invention, and FIG. 2 is a flowchart illustrating a moving object segmentation method of a compressed image according to an embodiment of the present invention. 3 is a detailed flowchart of the space-time division step shown in FIG.

Referring to FIG. 1, the moving object segmentation apparatus of the compressed image according to the present invention extracts a DC coefficient of a discrete cosine transform for each block by partially decoding a bit string of a compressed image in units of frames. A DC image reconstruction unit 1 for reconstructing a DC image reduced by 8 times each length, a DC image divider 2 for dividing a moving object using the reconstructed DC image, and partially decoding a bit string of a compressed image A motion vector extractor 3 for extracting a motion vector, and a space-time divider 4 for removing a wrongly extracted motion object from the motion object using the motion vector.

Here, the DC image segmentation unit 2 sets the DC image of the first access unit that compresses the first frame of the compressed image as a reference image, and the difference between the DC image of the other access unit that compresses another frame and the reference image. A differential image constructing unit constituting an image, a binary image constructing unit reconstructing a binary image composed of a motion pixel and a still pixel by lowpass filtering the difference image, and neighboring movement by median filtering the binary image It consists of an area divider for grouping pixels and dividing them into moving objects.

The operation and effects of the present invention configured as described above are as follows.

Referring to FIG. 2, the moving object segmentation method of the compressed image according to the present invention is as follows. First, in step S11, the compressed image is reconstructed into a DC image. That is, the DC coefficient of the discrete cosine transform is extracted for each block by partially decoding the compressed image in units of frames, and the image in one frame is reconstructed into a DC image reduced by 8 times in width and length.

Next, in step S12, the compressed image is partially decoded to extract a motion vector. A motion vector exists for each macroblock. Macroblocks that do not have predictive encoding are treated as having a motion vector of 0, and an I (Intra) picture is interpolated using motion vectors of two neighboring frames.

Next, the moving object is divided using the DC image reconstructed in steps S13 and S14. That is, in step S13, the DC image of the first access unit is set as a reference image, and the difference image between the DC image and the reference image of each access unit is obtained. Here, the access unit means one frame of compressed image bit string data.

In step S14, low pass filtering is performed on the difference image. In this case, the difference image is reconstructed into a binary image including a motion pixel having a pixel value greater than or equal to a threshold and a still pixel having a pixel value less than or equal to a threshold. Next, the median filtering of the binary image is performed. The neighboring motion pixels are grouped and divided into motion objects. At this time, the moving object area having a predetermined size or less is removed.

Next, in step S15, time-space division is performed for each moving object, which is performed for every access unit.

A detailed flow diagram for this space-time division is shown in FIG. 3.

First, when a motion object generated by DC image segmentation is input, the motion object region input in step S151 is divided using a motion vector. At this time, the two-dimensional translational motion model and the three-dimensional planar body motion model are divided into two areas. When the size of the small area among the divided two areas is more than 30% of the entire moving object area, the two areas in step S152 Split into areas. However, otherwise, keep this moving object as one moving object without splitting it.

In this way, if a motion object is divided by using a motion vector, when two objects having motions of different directions are extracted as one motion object from a DC image, the two motion objects are different because the directions of movement of the two objects are different. Can be separated. In addition, when the moving object and the shadow are extracted as one moving object, the moving object and the shadow may be separated. In other words, when the space-time segmentation technique is used, the segmentation performance is improved because the moving object by the DC image segmentation is more precisely segmented.

In step S153, the magnitude of the motion vector, the average and the variance of the difference image of the motion object are calculated for each motion object. That is, step S153 is performed for each moving object divided into two regions in step S152.

If the magnitude of the motion vector is smaller than the first threshold value, the average of the difference image is smaller than the second threshold value, and the variance of the difference image is smaller than the third threshold value in step S154, the process proceeds to step S155 to remove the motion object. After the steps S153 to S155 are executed for all the moving objects, the process ends.

In the moving object segmentation step using the DC image of FIG. 2, the object and the shadow may be divided into a single motion object by changing the brightness of some regions. In the space-time division step, the shadow area is removed after the object and the shadow are separated. do. In addition, in the moving object dividing step, objects having two different motions may be divided into one moving object. In the space-time dividing step, the two objects are divided into two regions to improve the segmentation performance.

As described above, according to the present invention, since the moving object can be directly partitioned without completely decoding the video bit stream of the MPEG-1,2 compressed surveillance environment, it is possible to save time and storage space and to reduce the calculation amount. Can be. In addition, since the spatio-temporal segmentation technique provides improved segmentation performance, object-based compressed image indexing and retrieval is possible.

Claims (8)

A DC image reconstruction unit for reconstructing a compressed image in a frame unit into a DC image using a DC coefficient of a discrete cosine transform for each block; A DC image splitter for dividing a moving object using the reconstructed DC image; A motion vector extractor for partially decoding the compressed image to extract a motion vector; and And a spatio-temporal divider for separating two adjacent motion objects using the motion vector and removing the wrongly extracted motion object. The method of claim 1, wherein the DC image segmentation unit, A difference image composition unit for obtaining a difference image between a DC image of one frame and a DC image of another frame; A binary image configuration unit which reconstructs the difference image into a binary image including a motion pixel and a still pixel by low pass filtering, and And a region dividing unit for performing median filtering on the binary image and grouping neighboring moving pixels to divide the moving image into a moving object. The apparatus of claim 2, wherein the space-time divider comprises: An area divider for dividing the motion object into two areas using a motion vector; A calculator for calculating the magnitude of the motion vector, the average and the variance of the difference image, for each of the moving objects, and And a segmentation performance improving unit for removing the motion object when the magnitude of the motion vector is smaller than the first threshold value, the mean of the difference image is smaller than the second threshold value, and the variance of the difference image is smaller than the third threshold value. Moving object segmentation apparatus for compressed video. A first step of partially decoding a compressed image in units of frames and reconstructing the DC image; A second step of extracting a motion vector by partially decoding the compressed image; A third step of dividing a moving object using the reconstructed DC image, and And dividing two adjacent motion objects by using the motion vector and removing the wrongly extracted motion object. The method of claim 4, wherein the third step, Setting a DC image of the first frame of the compressed image as a reference image; Obtaining a difference image between the DC image of another frame and the reference image; Low pass filtering the difference image to reconstruct a binary image including a motion pixel and a still pixel, and And performing median filtering on the binary image to group neighboring motion pixels to divide the motion image into motion objects. The method of claim 4 or 5, wherein the first step, And extracting the DC coefficient of the discrete cosine transform (DCT) for each block unit and reconstructing the compressed image into a DC image. The method of claim 4, wherein the fourth step, Dividing a motion vector extracted from the motion object; Dividing the motion object into two regions by using the motion vector; Calculating the magnitude and mean and variance of the motion vector and the difference image for each motion object, and If the size of the motion vector of each of the motion objects is smaller than the first threshold value, the average of the difference image is smaller than the second threshold value, and the variance of the difference image is smaller than the third threshold value, the motion object is removed to improve segmentation performance. Moving object segmentation method of a compressed image, characterized in that it comprises a step. On your computer, A first step of partially decoding a compressed image frame-by-frame, extracting a DC coefficient of a discrete cosine transform for each block, and reconstructing it into a reduced DC image; A second step of extracting a motion vector by partially decoding the compressed image; A third step of setting the DC image of the first frame of the compressed image as a reference image; A fourth step of obtaining a difference image between the DC image of another frame and the reference image; A fifth step of reconstructing the difference image into a binary image composed of a moving pixel and a still pixel by performing low pass filtering; A sixth step of grouping neighboring motion pixels by median filtering the binary image and dividing the binary image into a motion object; A seventh step of separating the motion object into two regions by using the motion vector; An eighth step of calculating a magnitude and a mean and variance of a motion vector and a difference image for each of the moving objects, and If the size of the motion vector of each of the moving objects is smaller than the first threshold value, the average of the difference image is smaller than the second threshold value, and the variance of the difference image is smaller than the third threshold value, the motion object is determined to be unnecessary and removed. And a computer-readable recording medium having recorded thereon a program for executing the ninth step of improving partition performance.