KR0160458B1 - Moving picture compressing device and method for estimation of motion vector with mask block matching method - Google Patents

Abstract

The present invention relates to a video compression apparatus and method, and provides a video compression apparatus and method for reducing the amount of transmission data by compressing only a portion where a prediction error is concentrated by using a mask block matching method that can predict a motion vector in a simpler manner. Motion estimating means (11) for estimating motion; Grid deforming means 12 for deforming the grid; Image synthesizing means (13) for synthesizing the images; Error image processing means (14) for outputting an error image; Image generating means (15) for generating a reproduced image; Image storage means (16) for storing an image; And a buffering means (17) for outputting a motion vector, a compression mode, and an error image to the outside. In the method applied to the apparatus, a prediction error is performed after synthesizing the current input image after modifying the motion estimation and grid. First step (21 to 25) to calculate; A second step (26, 28, 29) for performing intra coding; And a third step (27 to 29) of compressing using the motion vector and the error image, the motion vector estimation can be easily performed, and the amount of transmission data can be reduced.

Description

Motion picture compression apparatus and method for predicting motion vectors using mask block matching

1 is a block diagram of a video compression apparatus according to the present invention.

2 is a flow chart in accordance with the present invention.

3 is a diagram showing the shape of the grid overlaid image.

4 is a shape diagram of a grid deformed by a motion vector.

5 is a shape diagram of an image synthesized by a texture mapping method using only a modified grid and a previous frame image.

6 is a shape diagram of a difference image between a synthesized image and an original image.

7 is a diagram of a type of image reproduced according to the number of error blocks.

* Explanation of symbols for main parts of the drawings

11: motion estimation unit 12: grid deformation unit

13 image synthesis unit 14 error image processing unit

15: image generating unit 16: image storage unit

17: buffer

The present invention relates to a video compression apparatus and method for reducing a transmission data amount by predicting a motion vector using a mask block matching method.

Most existing image compression apparatuses are based on a block-based method of performing motion estimation and prediction by dividing an image into subblocks. The motion vectors of each block are predicted independently, and all pixel values in a block have the same motion vector.

Therefore, the movement such as rotation, enlargement, reduction, etc. of the object has a problem that it is difficult to sufficiently compensate by the conventional block-based method.

In order to solve this problem, the proposed texture mapping method using affine transform uses an iterative estimation method using gradient constraints for motion vector prediction.

However, the texture mapping method using the affine transformation also takes a lot of computation time and there is a problem that the accuracy of the motion vector can be dependent on the initial value of the motion vector.

In order to solve the above problems, the present invention compresses only a portion of the prediction error concentrated using a discrete cosine transform (DCT) using a mask block matching method that can predict a motion vector in a simpler manner. It is an object of the present invention to provide a video compression apparatus and method for reducing.

An apparatus of the present invention for achieving the above object comprises: motion estimation means for outputting a motion vector by estimating a motion between an input image currently input from the outside and a previous frame image; Grid deforming means for deforming a grid using a motion vector obtained by the motion estimating means; Image synthesis means for synthesizing the current input image by receiving the grid information deformed by the grid deforming means and the previous frame image; The difference between the input image and the synthesized image generated by the image synthesizing means is divided into subblocks to calculate an average error of each block, and selectively compresses a predetermined number in order from a block having a large average error value. Error image processing means for outputting an image and determining and outputting an extrusion mode; Image generation means (Image Gemeration) for generating a reproduced image by adding the error image output from the error image processing means and the composite image generated by the image synthesis means according to the compression mode output from the error image processing means; Image storage means for receiving the image reproduced by the image generating means and storing the reproduced image for outputting a previous frame image to the motion estimation means and the image combining means; And a buffering means for receiving a motion vector calculated by the motion estimation means, a compression mode determined by the error image processing means, and an error image and outputting them to the outside.

According to another aspect of the present invention, there is provided a method, comprising: motion estimation means for estimating motion between an input image currently input from the outside and a previous frame image to output a motion vector; Grid deforming means for deforming a grid using a motion vector obtained by the motion estimating means; Image synthesis means for synthesizing the current input image by receiving the grid information deformed by the grid deforming means and the previous frame image; The difference between the input image and the synthesized image generated by the image synthesizing means is divided into subblocks to calculate an average error of each block, and selectively compresses a predetermined number in order from a block having a large average error value. Error image processing means for outputting an image and determining and outputting a compression mode; Image generating means (Image Gemeration) for generating a reproduced image by adding an error image output from the error image processing means and a synthesized image generated by the image combining means in accordance with a compression mode output from the error image processing means; Image storage means for receiving the image reproduced by the image generating means and storing the reproduced image for outputting a previous frame image to the motion estimation means and the image combining means; And a buffering means for receiving a motion vector calculated by the motion estimating means, a compression mode determined by the image processing means, and an error image and outputting the result to the outside. A first step of synthesizing the current input image by a texture mapping method using the deformed grid and the previous frame image and then calculating a prediction error and setting a threshold difference to determine a compression mode; A second step of generating and storing a reproduced image after performing intra coding to compress the input image using only the information of the input image if the prediction error is greater than the threshold after performing the first step; And after performing the first step, if the prediction error is smaller than the threshold, a third step of compressing the current input image using the motion vector and the error image, and generating and storing the reproduced image.

Hereinafter, with reference to the accompanying drawings will be described an embodiment according to the present invention;

1 is a block diagram of a video compression apparatus according to the present invention, 11 is a motion estimation unit, 12 is a grid deformation unit, 13 is an image synthesizer, 14 is an error image processing unit, 15 is an image generator, 16 is an image storage unit And 17 represent buffers, respectively.

Looking at the specific configuration and operation, the motion estimation unit 11 synthesizes the motion vector by estimating the motion of the grid between the current input image from the outside and the previously compressed image output from the image storage unit 16 Output to section 13 and buffer 17. At this time, the motion of the grid is estimated by the following method.

A grid composed of 16x16 sub-blocks is overlaid on a previously captured image output from the image storage unit 16 (FIG. 3). A 15 × 15 mask is constructed around the intersections of the grids and then compared with the currently input image to find a block that minimizes the error between the masks. In this case, the search area has ± 8 pixels in the horizontal and vertical directions, respectively, and the block having the smallest error by calculating the difference between the mask block covered by the grid point on the previous frame image and the blocks in the search area of the current input image. Next, the difference between the x coordinate and the y coordinate between the block and the mask block is the motion vector.

In the grid deforming unit 12, the grids are transformed by moving the intersections of the grids using the motion vectors obtained by the motion estimation unit 11, and then connecting the grid intersections again (FIG. 4). Output to the image synthesizing unit 13.

The image synthesizing unit 13 synthesizes the current input image by using the modified grid information output from the grid deforming unit 12 and a previously compressed image output from the image storing unit 16 to convert the synthesized image into an error image processing unit. 14 and the image generating unit 15 are output. That is, since each triangle constituting the grid before and after deformation is one-to-one correspondence, a modified image can be obtained by a texture mapping method using affine transformation, and this image is synthesized by using the modified grid with the previous frame image. Becomes the current video. Texture mapping is a method in which all the pixel values in the triangle A'B'C 'are calculated using affine transformation from the pixel values in the triangle ABC when the triangle ABC is transformed into the triangle A'B'C'.

The error image processor 14 obtains a difference image between the input image and the synthesized image output from the image synthesizer 13, divides the difference image into 8 × 8 subblocks, and calculates an average value of each subblock. After rearranging in large order, only M blocks are compressed to output an error image to the image generating unit 15 and the buffer 17. At this time, the total error average value is calculated, and this value is compared with a threshold to determine a compression mode of the input image, and this value is output to the buffer 17 and the image generator 15.

The image generating unit 15 receives a composite image output from the image synthesizing unit 13 and an error image output from the error image processing unit 14, and generates a reproduced image according to a compression mode output from the error image processing unit 14. And output to the image generator 16. That is, when the compression mode is Intra, intra coding is performed on the input image while ignoring all the previous processes. When the compression mode is Inter, the image synthesis unit 13 The synthesized video synthesized at and the error image calculated by the error image processor 14 are added to generate a reproduced image, and the image is output to the image storage unit 15.

The buffer 17 receives the motion vector calculated by the motion estimation unit 11 and the compression mode and the error image determined by the error image processor 14, and transmits them to the outside.

The image storage unit 16 stores the playback image output from the image generating unit 15 for use as the previous frame image when the motion estimation unit 11 estimates the motion and the image combiner 13 synthesizes the current image.

2 is a flow chart according to the present invention.

In detail, when the input image is received (21), a grid composed of 16 × 16 basic blocks is overlaid on a previous frame image, and a 15 × 15 mask is covered around a grid intersection to be compared with the input image in a mask unit. In order to find the motion vector, the grid is transformed by the motion vector and overlaid on the current input image (22). In this case, the search area is set to ± 8 pixels in the horizontal and vertical directions, respectively, and the block having the smallest error is calculated by calculating the difference between the mask block covered by the grid point on the previous frame image and the blocks in the search area of the current input image. Next, the difference between the x coordinate and the y coordinate between the block and the mask block is the motion vector.

Thereafter, a composite image of the current input image is obtained using the modified grid and the previously compressed image (23). In this case, the composite image reproduction uses a texture mapping method using affine transformation. That is, since each triangle constituting the grid before and after the deformation is one-to-one correspondence, the deformed image can be obtained by the texture mapping method using affine transformation, and the image is synthesized by using the deformed grid with the previous frame image. It becomes the current video. Texture mapping is a method in which triangle ABC can calculate all pixel values in triangle A'B'C 'using affine transformation from pixel values in triangle ABC.

Next, after obtaining the difference image between the input image and the composite image price, the difference image is divided into 8 × 8 subblocks, the average value of each subblock is calculated, rearranged in the order of the larger values, and then compressed by only M blocks. The error is calculated and the threshold is set 25.

Then, the compression mode of the input image is determined by comparing the prediction error and the threshold (25). That is, if the prediction error is greater than the threshold, intra coding is performed using only the information of the input image to compress the input image (26), and then the reproduced image is generated (28) and stored (29). If the threshold value is smaller than the threshold, the current input image is compressed using the motion vector and the error image (27), and then the reproduced image is generated (28) and stored (29). In this case, in the intra coding mode, one of the methods used for conventional still image compression is compressed.

Looking at the process of compressing using the motion vector and the error image in detail, when looking at the difference image between the input image and the image synthesized using texture mapping (Fig. 6), the error is concentrated on a specific part such as the eye, nose, and mouth. Instead of compressing all of the error images, only the upper M blocks with large errors are selectively compressed (28). That is, even if only about one-tenth of the error blocks are selectively compressed, there is no significant effect on the quality of the reproduced video.

Subsequently, after generating the reproduced image by adding the compressed error image and the synthesized image synthesized by the texture mapping method (28), the reproduced image is stored in a frame store for use in compressing the next image (29).

In the experiment according to the present invention, a Common Intermediate Format (CIF) image of Miss 352 × 288 was used.

FIG. 3 is a shape diagram of an image overlaid with a grid, and is an image overlaid with a grid before deformation. FIG. 4 is a shape diagram of a grid deformed with a motion vector. 5 is a shape diagram of an image synthesized by the texture mapping method using only the modified grid and the previous frame image.

FIG. 6 is a shape diagram of the difference image between the synthesized image and the original image, and it can be seen that the prediction error is concentrated in the large part of the local motion such as the eyes, nose, and mouth.

FIG. 7 is a shape diagram of the image reproduced according to the number of error blocks. The experiment results are obtained by selectively compressing a block having a large error by using an error concentration phenomenon. The image quality is obtained by compressing only about 100 of the 584 error blocks. You can see that there is no big difference.

As described above, the present invention can easily perform motion vector estimation using a mask block matching method, reduce the amount of transmitted data by compressing only a portion where the prediction error is concentrated using a discrete cosine converter, and reduce an error block. Compression of only about 1/10 can produce a good quality video.

Claims (8)

Motion estimating means (11) for estimating motion between an input image currently input from the outside and a previous frame image to output a motion vector; Grid deformation means (12) for deforming a grid using a motion vector obtained by said motion estimation means (11); Image synthesis (13) for receiving the grid information modified by the grid modifying means (12) and the previous frame image to synthesize a current input image; The difference between the input image and the synthesized image generated by the image synthesizing means 13 is divided into subblocks, and the average error of each block is calculated to selectively select a predetermined number in order from blocks having the large average error value. Error image processing means (14) for compressing and outputting an error image, and for determining and outputting a compression mode; Image generation for generating a playback image by adding the error image output from the error image processing means 14 and the composite image generated by the image synthesizing means 13 in accordance with the compression mode output from the error image processing means 14. Means (Image Gemeration) 15; An image storage means for receiving the image reproduced by the image generating means 15 and storing the reproduced image for outputting a previous frame image to the motion estimating means 11 and the image synthesizing means 13; 16); And a masking means (17) for receiving the motion vector calculated by the motion estimation means (11), the compression mode determined by the error image processing means (14), and the error image, and outputting them to the outside. A video compression device that predicts motion vectors in a way. Motion estimating means (11) for estimating motion between an input image currently input from the outside and a previous frame image to output a motion vector; Grid deformation means (12) for deforming a grid using a motion vector obtained by said motion estimation means (11); Image synthesis means (13) for receiving the grid information transformed by the grid transformation means (12) and the previous frame image to synthesize a current input image; The difference between the input image and the synthesized image generated by the image synthesizing means 13 is divided into subblocks, and the average error of each block is calculated to selectively select a predetermined number in order from blocks having the large average error value. Error image processing means (14) for compressing and outputting an error image, and for determining and outputting a compression mode; Image generation for generating a playback image by adding the error image output from the error image processing means 14 and the composite image generated by the image synthesizing means 13 in accordance with the compression mode output from the error image processing means 14. Means (Image Gemeration) 15; An image storage means for receiving the image reproduced by the image generating means 15 and storing the reproduced image for outputting a previous frame image to the motion estimating means 11 and the image synthesizing means 13; 16); And a buffering means (17) for receiving a motion vector calculated by the motion estimation means (11), a compression mode determined by the error image processing means (14), and an error image, and outputting them to the outside. After receiving the input image and performing motion estimation and grid transformation, the current input image is synthesized using the texture mapping method using the deformed grid and the previous frame image, and then the prediction error is calculated and the threshold is compressed. First steps 21-25 to determine the mode; After performing the first steps 21 to 25, if the prediction error is larger than the threshold, the second step of generating and storing the reproduced image after performing intra coding to compress the input image using only the information of the input image. (26,28,29); And after performing the first steps 21 to 25, if the prediction error is smaller than the threshold, the third step (27 to 29) of compressing the current input image by using the motion vector and the error image, and then generating and storing the reproduced image. The video compression method for predicting the motion vector by a mask block matching method comprising a. The method of claim 2, wherein the motion estimation and the grid transformation of the first steps 21 to 25 are performed by overlaying a grid of 16 × 16 basic blocks on a previous frame image and then centering the grid intersections. A fourth step of finding a motion vector by covering a mask with a × 15 mask and comparing the input image with a mask unit; And a fifth step of modifying a grid by the motion vector and covering the current input image after performing the fourth step. The method of claim 3, wherein the searching of the motion vector in the fourth step comprises: a search block having ± 8 pixels in the horizontal and vertical directions, respectively, and a mask block and a current input image covered by a grid point on a previous frame image. The motion vector is a mask block matching method in which a block having the smallest error is found by calculating a difference between blocks in a search region of the block, and then the difference between the x and y coordinates between the block and the mask block is a motion vector. Video compression method to predict. The prediction error calculation process of the first steps 21 to 25 is performed by obtaining a difference image between the input image and the synthesized image, and then dividing the difference image into 8x8 subblocks. Calculating a mean value of the fourth step; And after performing the fourth step, rearranging the calculated average values in order of the greatest, and then compressing only a predetermined number of blocks to calculate a prediction error. The motion vector is predicted by a mask block matching method. Video compression method. 3. The method of claim 2, wherein the second steps 26, 28, and 29 are performed by performing intra coding to compress the input image using only the information of the input image when the prediction error is greater than a threshold value. 26); And a fifth step (28, 29) of storing a frame store for use in compressing a next video after generating the playback video after performing the fourth step 26. A video compression method for predicting a motion vector using a mask block matching method. 3. The method of claim 2, wherein the third step (27 to 29) comprises: a fourth step (27) of compressing the current input image using the motion vector and the error image when the prediction error is smaller than the threshold value; And after performing the fourth step 27, adding a compressed error image and a synthesized image synthesized by a texture mapping method to generate a reproduced image, and then using the reproduced image to compress the next image. And a fifth step (28, 29) of storing the motion vector compression method using the mask block matching method. The method of claim 7, wherein the compressing by using the motion vector and the error image in the fourth step 27 comprises: the difference between the image synthesized by using texture mapping and the input image is an error of eyes, nose, mouth, etc. A video compression method for predicting a motion vector by using a mask block matching method, which compresses only a predetermined number of blocks having a large error, rather than compressing all of an error image because it appears intensively in a specific part.