KR19980036116A - Variable Block Motion Estimation and Compensation Method Using Tree Erase Algorithm and Its Apparatus - Google Patents

Abstract

The present invention relates to a method and apparatus for motion compensation and encoding of a variable block, and a novel method and apparatus for estimation and compensation of variable block motion that can effectively reflect not only the overall characteristics of an image but also local characteristics. In order to provide an appropriate size of an image block for encoding and motion compensation, a quad tree formed by dividing an image block in a quad tree scheme to reflect local characteristics of the image and nodes having a minimum degradation ratio in the quad tree are provided. The present invention provides a method and apparatus for motion estimation and compensation of a variable block using an eliminating tree cancellation algorithm.

Description

Variable Block Motion Estimation and Compensation Method Using Tree Erase Algorithm and Its Apparatus

The present invention relates to a method and apparatus for motion estimation and encoding of a variable block.

Recently, various moving picture coding techniques have been proposed for efficiently transmitting or storing moving pictures, and standard algorithms such as MPEG have been adopted. Most of these moving picture coding methods are composed of a hybrid form in which a motion compensation method and a transform coding method are combined. That is, the motion compensation eliminates the redundancy of the moving image in the time axis, and the transform coding is applied to the motion compensated difference signal to remove the correlation in the spatial domain. The most important factor that determines the performance of an encoder in such a hybrid video encoding technique is a motion compensation device that removes redundancy in the time axis of the video. A typical moving picture coder estimates and compensates motion by using a motion estimation technique such as block matching in units of fixed size blocks. That is, the image is divided into blocks of size K × K, and motion of each block is estimated and compensated using a motion prediction technique such as block matching for blocks of size K × K. In general, however, the motion of an object in a moving picture is independent of each object, and exhibits very complex motion characteristics. Therefore, if the image is divided into fixed sized blocks, there is a possibility that objects having various complex motions are included in the block, which causes the motion estimation and compensation performance of each block to be very degraded.

As a method for solving the problems of the fixed block motion estimation and compensation method, variable block motion compensation techniques have been proposed to perform motion estimation and compensation by variably dividing the block size according to the complexity of the motion.

1 is a flowchart of a general variable block motion estimation method. First, an image is divided into N × N sized blocks and inputted. The motion is estimated using the block matching for the input N × N block and the motion compensation error is calculated (step 100). It is determined whether the exercise compensation error is larger than the predetermined limit error value E (step 110). If the motion compensation error is larger than the predetermined threshold error value in the step, the block is divided into four N / 2 × N / 2 blocks (step 120). In this step, if the motion compensation error is smaller than the predetermined threshold error value, the motion estimation is terminated. If the block is divided in step 120, it is determined whether the size of the divided block is larger than the predetermined limit block size value B (step 130). If the size of the block divided in the above step is larger than the predetermined limit block size value (B), the step of estimating the motion and calculating the compensation compensation error (step 100) and whether the motion compensation error is larger than the predetermined limit error value (E) If it is determined (step 110) and large, the step of dividing and dividing the block further (step 120) is repeated. The process is repeated until the size of the block divided by step 130 is smaller than the predetermined limit block size value (B). That is, in the conventional method, variable block motion estimation and compensation are performed while continuously dividing an N × N block into a quadtree. The most important thing in the quadtree type variable block motion estimation method is to determine whether to perform further partitioning for each block. That is, the limit error value E of the motion compensation error of each block is determined. In the conventional variable block motion compensation method, the limit error value E is determined by an empirical method based on experiments on various images. However, this empirical method cannot effectively reflect the overall characteristics and the local motion characteristics of the image, and thus has a limitation in the performance of motion compensation. Recently, in order to more effectively estimate the variable block motion, methods that consider not only the motion estimation error but also the amount of bits required for the motion vector have been proposed. In this method, the Lagrange coefficient is used to define a new comparison function that considers the motion error and the bit amount at the same time, and determines whether to partition each block using the change ratio of the comparison function according to the partitioning of each block. However, since this method also determines the Lagrange coefficient by experimental method, there is a drawback that the optimal motion estimation performance cannot be expected.

The present invention is to compensate for the shortcomings of the prior art as described above, and an object of the present invention is to improve the motion compensation performance in the coding / decoding algorithm of a moving picture of a hybrid type in which a motion compensation and a transform encoding scheme are combined. The present invention provides a novel method and apparatus for estimating and compensating for variable block motion that can effectively reflect local characteristics as well as local characteristics.

It is still another object of the present invention to set a proper size of an image block for encoding and motion compensation in order to reflect local characteristics of the image. The present invention provides a method and apparatus for estimating and compensating motion of a variable block using a tree erasing algorithm for removing minimum nodes.

1 is a flow chart of a general variable block motion estimation method;

2 is a diagram illustrating image segmentation for an input image;

3 is a diagram illustrating quad tree generation corresponding to the image segmentation shown in FIG. 2;

4 is a diagram showing a symmetric quad tree whose entire image is divided into blocks of size N / 16 × N / 16 starting from an N × N level;

FIG. 5 is a view illustrating an asymmetric quad tree generated by removing an intermediate node having a minimum degradation ratio shown in FIG. 4; FIG.

6 shows a finally generated asymmetric quad tree;

FIG. 7 is a diagram illustrating an image block corresponding to each terminal node of the asymmetric quad tree of FIG. 6;

8 is a block diagram of a variable block motion estimation and compensation device using a tree cancellation algorithm according to the present invention;

9 is a detailed configuration diagram of the tree eraser of FIG. 8;

Explanation of symbols on the main parts of the drawings

80: image segmentation unit 81: block division unit

82: quadtree configuration unit 83: block matching unit

84: tree eraser 85: encoder

86: motion compensation unit 90: node selection unit

91: subtree removing unit 92: deterioration ratio updating unit

93: repeat control unit

In order to achieve the object of the present invention as described above, the present invention in the motion estimation and encoding method of a variable block, after dividing the input image into N × N pixel size, each N × N block is a predetermined minimum size A first step of generating a symmetric quad tree by dividing into blocks of a quad tree format; A second step of selecting an intermediate node having a minimum degradation ratio in the quad tree generated in the first step; A third step of removing from the quad tree the subtree having the intermediate node selected in the second step as a root node; A fourth step of updating the deterioration ratios of upper nodes of the intermediate node removed in the third step; The sum of the degradation ratios of all the terminal nodes of the asymmetric quad tree generated in the third step is determined to be smaller than a predetermined reference value, and when it is small, the tree erasing is stopped. Otherwise, the second, third and fourth steps are canceled. A fifth step of repeating the process of step until the sum of degradation ratios of all terminal nodes of the generated asymmetric quad tree is less than a predetermined reference value; And if the sum of deterioration ratios of all terminal nodes of the asymmetric quad tree generated by the fifth step is less than a predetermined reference value, image segmentation is performed to encode and compensate for the image by the size corresponding to the terminal nodes of the generated asymmetric quad tree. A motion estimation and encoding method of a variable block using a tri erasure algorithm, characterized in that it comprises a sixth step.

In addition, to achieve the above object, the present invention provides an apparatus for motion estimation and encoding of a variable block, comprising: an image splitter for dividing an input image into blocks of size N × N; A block dividing unit dividing the N × N size block generated by the image dividing unit into a quadtree method up to a predetermined minimum size block; A quadtree configuration unit constituting a quadtree as a data structure of blocks divided by the block division unit; A block matching unit for calculating a motion estimation error for all nodes of the quadtree; By using the motion estimation error calculated by the block matching unit, the intermediate node having the minimum degradation ratio is selected and removed, the degradation ratio of the upper nodes of the removed intermediate node is updated, and the sum of the degradation ratios of all terminal nodes is a reference value. A tree eraser which repeats until a smaller quad tree is generated; An encoder for dividing and encoding an image into blocks having sizes corresponding to terminal nodes of the asymmetric quadtree generated by the tree eraser; And a motion compensation unit for compensating for motion by dividing an image into blocks having a size corresponding to a terminal node of an asymmetric quadtree generated by the tree erasing unit. Provided is an encoding device.

Hereinafter, with reference to the drawings will be described in detail a variable block motion estimation and compensation method and a preferred embodiment of the apparatus using a tri- erasing algorithm according to the present invention.

First, a first step of the method according to the present invention will be described in which the input image is divided into quad tree shapes.

FIG. 2 is a diagram illustrating image segmentation for an input image, and FIG. 3 is a diagram illustrating quad tree generation corresponding to the image segmentation illustrated in FIG. 2.

First, referring to FIG. 2, the input image is divided into N × N size blocks, and the divided N × N size blocks are quadtree-divided. That is, a block of size N × N is divided into four blocks of size N / 2 × N / 2. Each N / 2 × N / 2 size block is quadtree-divided, and the quadtree division is repeated until the block size becomes a predetermined minimum size (M × M). The quad tree corresponding to the quad tree division of the input image is shown in FIG. 3. The root node, which is the highest node of the generated quad tree, represents the entire input image, and the nodes connected to the root node correspond to blocks of size N × N. Four nodes are connected to a node corresponding to an N × N size block, which represents four N / 2 × N / 2 size blocks generated by quadtree dividing the N × N size block. Similarly, four nodes representing a block of size N / 4 × N / 4 are connected to a node of a block of size N / 2 × N / 2. The nodes at the end of the quad tree represent blocks of size M × M, which is a predetermined minimum size. The nodes with the lowest level of the tree structure are called terminal nodes, and the nodes between the root node and the terminal nodes are called intermediate nodes.

As shown in FIG. 3, the quad tree first generated by quadtree dividing the input image is symmetrical.

The second step of the method according to the invention is the step of selecting an intermediate node having a minimum degradation ratio by a tree cancellation algorithm.

For all intermediate nodes of the quad tree generated in the first step, as shown in Equation 1 below, an intermediate node having a minimum deterioration ratio is selected.

[Equation 1]

In Equation 1, E _n represents a motion compensation error of a block corresponding to node n, and R _n represents a bit amount of a motion vector at this time. Is the motion compensation error when the motion compensation is obtained by dividing this block into a subtree connected to node n, Denotes the bit amount for the motion vectors at this time. In other words Is the sum of the motion compensation errors at the terminal nodes of the subtree with node n as the root node, Is defined as the sum of the motion vectors at this time. Referring to Fig. 2, for example, if the degradation ratio is calculated for an intermediate node representing a specific N / 2 × N / 2 size block, E _n is a motion compensation of N / 2 × N / 2 block. Exercise compensation error when Denotes the sum of the motion compensation errors when the motion compensation is performed with M × M blocks with respect to the terminal nodes of the subtree whose root node is the intermediate node representing the specific N / 2 × N / 2 size blocks. That is, it represents the sum of the motion compensation errors of each divided block when the motion compensation is performed by dividing a specific N / 2 × N / 2 size block into several M × M size blocks. Fig. 4 shows a symmetric quad tree starting from an N × N level, and shows an intermediate node having a minimum degradation ratio in a quad tree generated when the entire image is divided into blocks of size N / 16 × N / 16. It is displaying.

The third step of the method according to the present invention is to generate an asymmetric quad tree by removing the intermediate node having the minimum degradation ratio selected in the second step from the quad tree. FIG. 5 shows that the intermediate node with the degradation ratio indicated in FIG. 4 is removed. As shown in Fig. 5, removing an intermediate node having a minimum degradation ratio means removing a subtree whose root node is an intermediate node having a minimum degradation ratio.

The fourth step of the present invention is the step of updating the degradation ratio for all upper nodes having the intermediate node as the terminal node in the asymmetric quad tree from which the intermediate node is removed in the third step. In the asymmetric quad tree shown in FIG. 5, intermediate nodes whose degradation ratios are updated are indicated.

In the fifth step of the present invention, the deterioration ratios of all terminal nodes of the generated asymmetric quad tree are summed and compared with a preset reference value. If the sum of deterioration ratios of all terminal nodes of the asymmetric quad tree is smaller than the preset reference value, the motion compensation is performed with blocks of the size corresponding to each terminal node of the asymmetric quad tree, and conversely, If the sum of the deterioration ratios is greater than the preset reference value, after the deterioration ratio is updated by the fourth step, the intermediate node having the minimum deterioration ratio, which is the second stage, is selected for the remaining intermediate nodes, and the third stage is deterioration. A process of updating the degradation rate of the upper nodes of the removed intermediate node, the fourth step, and the degradation rate of all terminal nodes of the asymmetric quad tree, the fifth step Comparing the sum with a preset reference value, all terminal nodes of the asymmetric quad tree generated by the determination of the fifth step The sum of the group hwabi repeated until less than a predetermined threshold.

In the sixth step of the present invention, if it is determined that the sum of degradation ratios of all terminal nodes of the asymmetric quad tree generated as a result of the determination in the fifth step is smaller than a preset value, the sixth step corresponds to each terminal node of the generated asymmetric quad tree. The motion compensation is performed on the image blocks. FIG. 6 shows a final asymmetric quad tree, and FIG. 7 shows an image block corresponding to each terminal node of the asymmetric quad tree of FIG. As shown in Fig. 6, the final terminal nodes are composed of one N / 2 × N / 2 sized block, two N / 4 × N / 4 sized blocks and 40 N / 16 × N / 16 sized blocks. When the node corresponds to the block, the divided N × N block of the input image is divided into blocks as shown in FIG. 7 and motion compensated.

An apparatus for implementing the method according to the present invention as described above is described.

8 is a block diagram of a variable block motion estimation and compensation device using a tree cancellation algorithm according to the present invention.

As shown in the drawing, the apparatus according to the present invention divides an input image into blocks having an N × N size, and a predetermined minimum size of an N × N size block generated by the image splitter. A block splitting unit 81 for dividing a block into quadtrees, a quadtree configuring unit 82 constituting a quadtree as a data structure of blocks divided by the block splitting unit, and all the nodes of the quadtree. A block matching unit 83 that calculates a motion estimation error with respect to the block, and selects and removes a middle node having a minimum degradation ratio by using the motion estimation error calculated by the block matching unit, and deteriorates the upper nodes of the removed intermediate nodes. Is a tree eraser 84 which repeats and repeats this until a sum of degradation ratios of all terminal nodes is smaller than a reference value, the asymmetric quad generated by the tree eraser. The image is divided into blocks of the size corresponding to the terminal node of the asymmetric quadtree generated by the encoder 85 and the tree eraser 84. It comprises a motion compensation unit 86 to divide and compensate for the exercise.

9 is a detailed configuration diagram of the tree eraser of FIG. 8.

In addition, the tree eraser 84 selects an intermediate node having a minimum degradation ratio using the motion estimation error calculated by the block matching unit 82, and the node selector selected by the node selector. Subtree removal unit 91 for generating an asymmetric tree by removing a subtree as a node as a root node, and a degradation ratio updating unit 92 for updating the degradation ratio of upper nodes of nodes selected by the node selection unit 90. ) And the deterioration ratios of all terminal nodes of the asymmetric tree whose deterioration ratios are updated by the deterioration ratio updating unit, and as a result, stops the tree erasing if it is smaller than the reference value and outputs the asymmetric tree generated by the subtree removing unit. Otherwise, the node selector causes the node selector to repeat selecting the intermediate node having the lowest degradation ratio in the asymmetric tree whose degradation ratio is updated. It consists of 93.

Hereinafter, the operation of each unit of the variable block motion estimation and encoding apparatus using the tree cancellation algorithm according to the present invention will be described in more detail.

The image splitter 80 divides the input image to be currently encoded in a moving picture into blocks of size N × N. The block dividing unit 81 divides an N × N size block input from the image dividing unit into a quadtree shape up to a predetermined minimum size. That is, an N × N block is divided into four N / 2 × N / 2 blocks, and each N / 2 × N / 2 block is divided into four N / 4 × N / 4 blocks. The process of dividing the image into blocks is performed up to a block having a predetermined size of M × M pixels. The quad tree construction unit 82 constitutes a symmetric quad tree as a data structure of blocks divided by the block division unit 81. An example of a symmetric quad tree generated by dividing an N × N block into N / 16 × N / 16 size blocks in a quad tree manner is shown in FIG. The block matching unit 82 calculates a motion estimation error for all nodes of the quad tree configured by the quad tree construction unit 82. That is, the motion compensation error is calculated when the motion compensation is performed by dividing the block in the form of a subtree connected to each node for each node. In other words, for the node corresponding to the size of N / 2 × N / 2, the motion compensation error when the motion compensation is performed with the image block of size N / 2 × N / 2 is calculated. The node selector 90 of the tree eraser 84 selects an intermediate node having a minimum degradation ratio by using the motion estimation error calculated by the block matcher 82. The degradation ratio of each node is as defined in Equation 1 above. The subtree remover 91 of the tree eraser 84 removes a subtree whose root node is the node selected by the node selector 90 to generate an asymmetric tree. The output of the asymmetric tree by the subtree removing unit 91 is controlled by the repeating control unit 93. The degradation ratio updating unit 92 of the tree eraser 84 updates the degradation ratios of the upper nodes of the nodes selected by the node selector 90. The node selected by the node selector 90 needs to update the deterioration ratio value because the nodes of the lower level have been removed by the subtree remover 91. The iteration control unit 93 of the tree erasing unit 84 sums the deterioration ratios of all terminal nodes of the asymmetric tree whose deterioration ratio is updated by the deterioration ratio updating unit 92, and as a result, stops the tree erasing if it is smaller than the reference value. Outputting the asymmetric tree generated by the subtree removing unit 91; otherwise, causing the node selecting unit 90 to repeat selecting the intermediate node having the minimum deterioration ratio in the asymmetric tree whose deterioration ratio is updated. It consists of a repeating control unit 93.

As described above, the motion estimation and encoding method of a variable block using the present invention's tree erasing algorithm, and the apparatus according to the present invention, locally determine the degradation ratio of an image in determining the size of a block for motion estimation and encoding of a video signal. Since the size of the variable block is determined according to the deterioration ratio, it provides a more effective method in encoding and motion estimation of a moving image.

Claims (5)

In the motion estimation and encoding method of a variable block, A first step of dividing the input image into N × N pixel sizes and dividing each N × N block into a quad tree format to a predetermined minimum size block to generate a symmetric quad tree; A second step of selecting an intermediate node having a minimum degradation ratio in the quad tree generated in the first step; A third step of removing from the quad tree the subtree having the intermediate node selected in the second step as a root node; A fourth step of updating the deterioration ratios of upper nodes of the intermediate node removed in the third step; The sum of the degradation ratios of all the terminal nodes of the asymmetric quad tree generated in the third step is determined to be smaller than a predetermined reference value, and when it is small, the tree erasing is stopped. Otherwise, the second, third and fourth steps are canceled. A fifth step of repeating the process of step until the sum of degradation ratios of all terminal nodes of the generated asymmetric quad tree is less than a predetermined reference value; And If the sum of deterioration ratios of all terminal nodes of the asymmetric quadtree generated by the fifth step is less than a predetermined reference value, the image splitting is performed to encode and motion-compensate the size corresponding to the terminal nodes of the generated asymmetric quadtree. Motion estimation and encoding method of a variable block using a tree erase algorithm, characterized in that it comprises a six steps. The method of claim 1, wherein the degradation ratio is defined by the following equation in the second step of selecting an intermediate node having a minimum degradation ratio: Here, E _n represents the motion compensation error of the block corresponding to the node n, R _n represents the bit amount for the motion vector at this time, Is the motion compensation error when the motion compensation is obtained by dividing this block into a subtree connected to node n, Denotes the bit amount for the motion vectors at this time. In the motion estimation and encoding apparatus of a variable block, An image splitter dividing the input image into blocks of size N × N; A block dividing unit dividing the N × N size block generated by the image dividing unit into a quadtree method up to a predetermined minimum size block; A quadtree configuration unit constituting a quadtree as a data structure of blocks divided by the block division unit; A block matching unit for calculating a motion estimation error for all nodes of the quadtree; By using the motion estimation error calculated by the block matching unit, the intermediate node having the minimum degradation ratio is selected and removed, the degradation ratio of the upper nodes of the removed intermediate node is updated, and the sum of the degradation ratios of all terminal nodes is a reference value. A tree eraser which repeats until a smaller quad tree is generated; An encoder for dividing and encoding an image into blocks having sizes corresponding to terminal nodes of the asymmetric quadtree generated by the tree eraser; And Motion compensation and coding of a variable block using a tree erase algorithm, characterized in that it comprises a motion compensation unit for compensating the motion by dividing the image into blocks of the size corresponding to the terminal node of the asymmetric quadtree generated by the tree eraser Device. The method of claim 3, wherein the tree eraser A node selecting unit which selects an intermediate node having a minimum degradation ratio using the motion estimation error calculated by the block matching unit; A subtree removing unit generating an asymmetric tree by removing the subtree having the node selected by the node selecting unit as the root node; A deterioration ratio updating unit for updating deterioration ratios of upper nodes of the node selected by the node selecting unit; And The deterioration ratio of all terminal nodes of the asymmetric tree whose deterioration ratio has been updated by the deterioration ratio updating unit is summed, and as a result, when it is less than the reference value, the tree is stopped and the asymmetric tree generated by the subtree removing unit is output. Otherwise, the node selector comprises an iterative control unit for repeating the selection of the intermediate node having the minimum degradation ratio in the asymmetric tree whose degradation ratio has been updated. 5. The apparatus of claim 4, wherein the node selector selects a node having a minimum deterioration ratio defined by the following equation: Here, E _n represents the motion compensation error of the block corresponding to the node n, R _n represents the bit amount for the motion vector at this time, Is the motion compensation error when the motion compensation is obtained by dividing this block into a subtree connected to node n, Denotes the bit amount for the motion vectors at this time.