KR0134489B1 - Adaptive motion compensation apparatus and method therefor - Google Patents

Abstract

The present invention relates to an adaptive motion compensation image encoding apparatus. The present invention relates to inputting first input image data from a frame to be currently encoded and second input image data from a previous frame previously encoded and stored in a memory. And a motion classifier for classifying blocks in which a large motion has occurred, wherein the motion classifier outputs the motion classification information output from the motion classifier, image data of a current frame input for encoding, and image data of a previous frame stored in the memory. The motion vector information is extracted using the motion vector information, and the motion compensation prediction is performed using the motion vector information and the image data of the previous frame. Therefore, by performing motion compensation in consideration of a region where a large motion occurs, accurate motion information can be extracted and image compression can be performed more efficiently.

Description

Adaptive Motion Compensation Image Coding Device

1 is a block diagram illustrating a conventional motion compensated video encoding apparatus including motion compensated prediction and discrete cosine transform.

2 is a diagram illustrating an example of extracting motion information.

3 is a block diagram constructed according to an embodiment of a motion compensation video encoding apparatus according to the present invention.

4 is a detailed block diagram of the motion classifier shown in FIG.

5 is a graph illustrating MC / no MC decision rules.

6 is a diagram for explaining an MC / no MC region.

7 is a diagram for explaining measurement of a displacement block difference signal in a search region.

* Explanation of symbols for main parts of the drawings

10 subtractor 12 discrete cosine converter

14 quantizer 16 inverse quantizer

18: Inverse Discrete Cosine Converter 20: Adder

22: memory 24: motion compensated predictor

26: motion estimator 28: motion classifier

281: Area Divider 283: Large Motion Area Extractor

The present invention relates to a video encoding apparatus, and more particularly, to an adaptive motion compensation video encoding apparatus for more efficiently compressing data by applying a motion compensation prediction method in response to a motion movement amount of an inter-image moving object.

Recently, thanks to the remarkable development of digital video technology, it is possible to use digital video compression technology for application in various electronic communication fields such as video conference, digital broadcasting codec, video phone, etc. It became.

FIG. 1 is a block diagram illustrating a conventional video encoding apparatus including a motion compensated prediction (MCP) and a discrete cosine transform (DCT). In the figure, it is assumed that the input signal is PCM data having a level of 8 bits, that is, 0 to 255.

First, the input signal is divided into macro blocks of predetermined size, for example, 16 × 16 pixel blocks, and motion vector information is extracted for each block, and each block is again sub-blocked. For example, motion compensation prediction and discrete cosine transform are performed by dividing into 8x8 pixel blocks.

In this case, the motion vector information of the subblocks in the giant block is estimated to be the same as the motion vector information of the giant block and used for motion compensation prediction. The performance of motion compensation prediction is to compress the image by eliminating the inter-frame redundancy by predicting the movement of the Inter Frame object, and the discrete cosine transform uses the energy concentration according to the frequency of the pixels in each sub-block. In other words, image compression is achieved by removing intra frame redundancy.

However, in the case of such motion compensation prediction, motion vector information detected in units of large blocks is used as described above. In this case, it is assumed that motion between frames is not very large. However, if there is a large motion, accurate motion is detected. It is impossible to do.

That is, for example, as shown in FIG. 2, a block a of the current frame t indicates a current block from which motion information is to be extracted. In addition, the dotted rectangle represents a search area in the previous frame t-1 for extracting motion information, and the block b in the search area of the previous frame t-1 represents the current frame t. It represents one candidate block having the same spatial position as the block (a) of. In addition, another candidate block c in the search region of the previous frame t-1 represents a block moved from the current frame t to the position of block a. That is, when the candidate block c is in the search region, the motion vector information can be extracted relatively accurately. However, when a large motion outside the search region is generated as shown in the figure, accurate motion information cannot be extracted. There was a problem that the efficiency is lowered.

Accordingly, an object of the present invention is to provide an adaptive motion compensation video encoding apparatus for performing motion compensation prediction in consideration of a region where a large motion occurs in order to solve the above problem.

In order to achieve the above object, the adaptive video encoding apparatus includes: a subtractor for generating difference image data by subtracting image data from which motion compensation prediction is performed on image data of a block in a current frame to be encoded; A discrete cosine converter for converting the difference image data output from the subtractor into a transform coefficient of a frequency domain to remove redundancy on a spatial domain; A quantizer for quantizing the discrete cosine transformed transform coefficients from the discrete cosine transformer to a predetermined quantization step size; An inverse quantizer for reconstructing image data quantized by the quantizer to transform coefficients before quantization; An inverse discrete cosine transformer for restoring inverse quantized transform coefficients through the inverse quantizer into block pixel data on a spatial domain before the discrete cosine transform; An adder for adding spatial domain transformed block pixel data and a result of performing motion compensation prediction through the inverse discrete cosine transformer; A memory for storing the image data output through the adder as a previous frame; A motion classifier for classifying blocks in which large motions are generated by inputting image data from the current frame and image data from a previous frame stored in the memory; A motion estimator for extracting motion vector information using motion classification information output from the motion classifier, image data of the current frame and image data of a previous frame stored in the memory; A motion compensation predictor for performing a motion using the motion vector information output from the motion estimator and the image data of a previous frame stored in the memory and applying the motion to the subtractor and the adder, wherein the motion compensation prediction classifier Displaced Block Difference (DBD) signals representing respective movement displacements between the blocks of the current frame and the plurality of candidate blocks in the search region of the previous frame are obtained, their average values are calculated, and the average values and the respective values are calculated. The block difference signal BD between the block of the frame to be currently encoded and the previous frame block at a position spatially equal to the condition where the difference value between the displacement block difference signal DBD is smaller than the first threshold value T1. If satisfies a condition larger than the second threshold value T2, it is determined that the large motion generation area, Group the second threshold value (T2) is characterized in that the larger value than zero.

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

3 is a block diagram of an adaptive motion compensation video encoding apparatus according to an embodiment of the present invention. The encoding device is configured to subtract a result of performing motion compensation prediction on image data of a current frame input for encoding. A subtractor 10, a discrete cosine transformer 12 for converting image data of a predetermined block output through the subtractor 10 into a frequency domain to remove data redundancy of the spatial domain, and a discrete cosine transformer 12 A quantizer 14 for quantizing the transform coefficients transformed through < RTI ID = 0.0 >) < / RTI > to a predetermined quantization step size, and also before transforming the quantized image data through the quantizer 14 Inverse quantizer (16) and inverse quantizer (16) for restoring the dequantized transform coefficients before discrete cosine transform An inverse discrete cosine transformer 18 for reconstructing the predetermined block pixel data on the spatial domain, and an inverse discrete cosine transformed image data through the inverse discrete cosine transformer 18 and a result value of performing motion compensation prediction An adder 20, a memory 22 for storing image data of a current frame output through the adder 20, a first input image data from a frame to be currently encoded, and previously encoded memory 22 A motion classifier 28 for classifying a block in which a large motion occurs by inputting second input image data from a previous frame stored in the second frame, motion classification information output from the motion classifier 28, A motion vector is generated using the image data of the current frame input to be encoded and the image data of the previous frame stored in the memory 22. The motion estimator 26 extracts the information, the motion vector information output from the motion estimator 26, and the image data of the previous frame stored in the memory 22 to perform motion compensation prediction to subtract the information. Motion Compensated Predictor 24 is applied to each adder 20.

4 is a detailed block diagram of the motion classifier shown in FIG.

4 is a block diagram of a motion compensation (MC) region and a motion compensation (NO) image according to an MC / no MC decision rule recommended by CCITT using current frame image data and previous frame image data input for encoding. And a large motion area extractor 283 for extracting an area where large motion has occurred with respect to the signal output from the area divider 281.

The operation of the present invention will now be described with reference to the accompanying drawings.

Referring to FIG. 3, the subtractor 10 subtracts the result of performing the motion compensation prediction to the image data of the current frame input to be encoded and outputs the result to the discrete cosine transformer 12. The discrete cosine transformer 12 converts image data of a predetermined block, for example, 8x8 pixel block, outputted through the subtractor 10 into a frequency domain to perform data compression to remove redundancy of the spatial domain. Output to (14). The quantizer 14 quantizes a predetermined transform coefficient transformed through the discrete cosine transformer 12 to a quantization step size, which is a predetermined palmification parameter, and outputs it to the next variable length coder (not shown). Output to the instrument 16. The inverse quantizer 16 restores the quantized image data through the quantizer 14 to the transform coefficients before being quantized and outputs them to the inverse discrete cosine transformer 18. The inverse discrete cosine transformer 18 restores the dequantized transform coefficients through the inverse quantizer 16 to a predetermined block in the spatial domain before the discrete cosine transform and outputs them to the adder 20. The adder 20 adds the image data of the predetermined block, which is inverse discrete cosine transformed by the inverse discrete cosine transformer 18, and a result of performing motion compensation prediction, and stores the result in the memory 22.

The motion classifier 28 receives the first input image data from the frame to be currently encoded and the second input image data from the previous frame which is previously data compressed and stored in the memory 22 to generate a large motion. Extract The motion classifier 28 will be described in detail with reference to FIG.

In FIG. 4, the motion classifier 28 is composed of an area divider 281 and a large motion area extractor 283. As shown in FIG. The area divider 281 divides the image into a motion compensation area MC and a no motion compensation area no MC according to the MC / no MC decision rule recommended by CCITT. The MC / no MC decision rule is based on the graph shown in FIG. The large motion region extractor 283 is for extracting a region in which large motion has occurred with respect to the output signal from the region divider 281. The characteristics when large motion occurs in the area divided by the area divider 281 are as follows.

First, it is impossible to estimate the exact motion with Block Matching Algorithm using small search area. Second, the distribution of Displaced Block Difference (DBD) in small search area is irregular due to the first feature. Third, a block difference signal (or block difference: BD) indicating a difference between a block of the frame to be currently encoded and a candidate block of a previous frame having the same spatial position is a large value. . Where DBD is the difference between the large blocks of the displacement frame obtained from the original and reconstructed signals.

In the present invention, only large motion is considered in the motion area, and block positions of the large motion are MC areas in which the shaded (peripheral) areas are MC areas, and blank (self) areas are no MC areas as shown in FIG. Appears to be.

In FIG. 6, when the blocks A, B, C, and D satisfy the above-mentioned three conditions corresponding to the large motion generation features, they are classified into a large motion area. Therefore, a mathematical relationship is derived from the three characteristics described above.

First, as shown in FIG. 6, the empty area is a no MC area, and when three or more of the four blocks above, below, left and right adjacent to the blank area are MC areas, irregularity or randomness investigation of the DBD is performed. That is, in the first step of the BMA, the investigation is performed as in FIG.

7 shows a search position in the search area, where '0' indicates a spatially same block position, and the remaining block positions indicate a position corresponding to the first stage in the three-stage BMA.

In this case, DBDs of the blocks of the current frame at position '0' and the candidate blocks of the previous frame at positions other than '0' are respectively represented by DBD (1), DBD (2),... It is represented by DBD (8). Here, DBD (i) (i = 1, ..., 8) may be a mean absolute error or a mean square error.

In this case, the average value A_DBD may be calculated by the following equation (1).

In addition, the difference value D_DBD (i) between the average value A_DBD and the individual DBD (i) is calculated by the following equation (2).

Then, the non-constantness of DBD (i) is determined from the following equation (3).

Where T1 represents a threshold.

That is, when the error between the individual DBD (i) value and the average value A_DBD is less than or equal to the predetermined threshold value T1 from Equation (3), it is determined that the distribution of the DBD (i) is irregular.

In the second step, an irregularity is determined from the size of the inter-block error BD at the same position spatially from the third feature, which is defined by the following equation (4).

That is, the block error BD is larger than the preset threshold T2. Where T2 has a value greater than zero. Therefore, in the present invention, a block satisfying both equations (3) and (4) is classified into a large motion area, and the classification information is applied to the motion estimator 26.

The motion estimator 26 estimates the motion vector information by using the motion classification information input from the motion classifier 28, the image data of the current frame input to be encoded, and the image data of the previous frame stored in the memory 22. Therefore, the motion predictor can be more accurately predicted by the motion predictor.

The motion compensation predictor 24 performs motion compensation prediction using the motion vector information output from the motion estimator 26 and the image data of the previous frame stored in the memory 22, and subtracts the result from the subtractor 10 and the adder. Deliver to 20.

As described above, in the adaptive motion compensation video encoding method and apparatus according to the present invention, by performing motion compensation prediction in consideration of a region where a large motion has occurred, extracting accurate motion vector information, it is possible to compress image data more efficiently. There is an advantage.

Claims (3)

An image encoding apparatus, comprising: a subtractor for generating difference image data by subtracting image data from which motion compensation prediction is performed on image data of a block in a current frame to be encoded; A discrete cosine transformer for converting the difference image data output from the subtractor into a transform coefficient of a frequency domain to remove redundancy on a spatial domain; A quantizer for quantizing the discrete cosine transformed transform coefficients from the discrete cosine transformer to a predetermined quantization step size; An inverse quantizer for restoring image data quantized by the quantizer to a transform coefficient before quantization; An inverse discrete cosine transformer for restoring inverse quantized transform coefficients through the inverse quantizer into block pixel data on a spatial domain before the discrete cosine transform; An adder for adding spatial domain transformed block pixel data and a result of performing motion compensation prediction through the inverse discrete cosine transformer; A memory for storing the image data output through the adder as a previous frame; A motion classifier for classifying a block in which a large motion occurs by inputting image data from the current frame and image data from a previous frame stored in the memory; A motion estimator for extracting motion vector information using motion classification information output from the motion classifier, image data of the current frame and image data of a previous frame stored in the memory; And a motion compensation predictor for performing motion compensation prediction using the motion vector information output from the motion estimator and the image data of a previous frame stored in the memory and applying the motion compensation prediction to the subtractor and the adder. Motion compensation video encoding device. The motion classifier of claim 1, wherein the motion classifier uses the current frame image data and the previous frame image data to display an image according to an MC / no NC decision rule recommended by CCITT. no area divider for MC); And a large motion region extractor for extracting the large motion generating region with respect to the image signal divided by the region divider. 3. The apparatus of claim 2, wherein the large motion region extractor obtains a displacement block difference signal (DDBD) representing each movement displacement between a block of the current frame and a plurality of candidate blocks in the search region of the previous frame. A block of the frame to be currently encoded at a position spatially equal to a condition that the average value thereof is calculated and a difference value between the average value and each of the displacement block difference signals DBD is smaller than a first threshold value T1; If the block difference signal BD between the control unit and the previous frame block satisfies a condition larger than the second threshold value T2, it is determined as the large motion generation region, and the second threshold value T2 is greater than zero. An adaptive motion compensation video encoding device.