KR100318471B1 - Apparatus and method for block matching motion estimation using table lookup vector quantization - Google Patents

Abstract

1. TECHNICAL FIELD OF THE INVENTION

The present invention relates to an apparatus and method for block matching motion estimation using table lookup vector quantization.

2. The technical problem to be solved by the invention

In order to improve the motion estimation speed, the present invention uses a table lookup vector quantization to calculate a difference between subblocks in advance and store them in a memory, and then estimate fast block matched motion vectors using calculated values stored in the memory. The present invention provides a block matched motion estimation apparatus and a method thereof.

3. Summary of Solution to Invention

The present invention provides a block-matched motion estimation method applied to a block-matched motion estimation apparatus, comprising: a first block for dividing a block of a previous frame and a block of a current frame to be motion estimated into a plurality of subblocks in a motion estimation search region; step; A second step of storing a vector quantization index by quantizing a table lookup vector of each subblock in the block of the previous frame in the divided search region; A third step of storing a vector quantization index by quantizing a sub-blocks in a block of the current frame by a table lookup vector; A fourth step of reading difference values between subblocks of a subblock in the block of the previous frame and a subblock in the block of the current frame and summing the difference values to calculate an interblock difference; And a fifth step of finding the smallest inter-block difference among the calculated inter-block differences of the current frame and the previous frame and transmitting the corresponding spatial position difference.

4. Important uses of the invention

The present invention is used for motion estimation and the like.

Description

Apparatus and method for block matching motion estimation using table lookup vector quantization

The present invention relates to a block matched motion estimation apparatus and method using table lookup vector quantization in order to increase the speed of the motion estimator.

1 is an explanatory diagram showing a conventional method for calculating the difference between two blocks.

As shown in the figure, in the conventional motion estimation method, a sum of absolute value (hereinafter referred to as "SAD") of the difference between each pixel or the difference between each pixel in order to calculate the difference between two blocks. Because of the sum squared difference (hereinafter referred to as "SSD"), a large amount of computation is required. For example, if a block is composed of L lines and L pixels (i.e., the block size is L × L), the conventional method using SAD uses L ² subtraction operations, L to calculate the difference between the two blocks. ^2-1 requires a single addition operation and comparison operation times L ² is, in the conventional method using the SSD L ² is a single multiply operation and add operations L ^2-1 times are required.

Further, when the motion estimation search region is horizontally + -I pixels, vertically + -J pixels, and performs integer-accuracy motion estimation (i.e., the number of candidate motion vectors is (2x). In the case of I + 1) × (2 × J + 1), in order to obtain the SAD in each candidate motion vector, approximately (2 × I + 1) × (2 × J + 1) additions per pixel, respectively, Subtraction and comparison operations are required, and (2 × I + 1) × (2 × J + 1) −1 comparison operations are required per block to find an optimal motion vector among candidate motion vectors. For example, if the size of the motion estimation unit block is 16 × 16, the motion estimation search area is + -16 horizontally, and + -16 vertically, the number of additions and subtractions is approximately 1089 times per pixel and 1093 comparison operations. This is necessary.

Therefore, in the conventional motion estimation method, since the SAD between each pixel and the SSD between each pixel are used to calculate the difference between the two blocks, a large amount of computation is required, and thus there is a problem in that a low cost real-time implementation has many limitations.

The present invention has been proposed to solve the above problems. In order to improve the motion estimation speed, the difference between the subblocks is pre-calculated using the table lookup vector quantization and stored in a memory, and then stored in the memory during motion estimation. An object of the present invention is to provide a block matched motion estimation apparatus and a method for estimating a fast block matched motion vector using a calculated value.

1 is an explanatory diagram showing a conventional method for calculating the difference between two blocks.

2 is a general block matching motion estimation section.

3 is a block diagram of a block matching motion estimation apparatus using table lookup vector quantization according to an embodiment of the present invention.

Figure 4 is an embodiment explanatory diagram showing a difference calculation method between two blocks according to the present invention.

5 is a diagram illustrating a relationship between a block that is a motion estimation unit and a subblock that is a vector quantization unit, according to an embodiment of the present invention.

6 is a diagram illustrating an embodiment of hierarchical vector quantization of each subblock according to the present invention;

7 is a flowchart illustrating a block matched motion estimation method using table lookup vector quantization according to the present invention.

* Explanation of symbols for main parts of the drawings

301: address generator 302: first frame memory

303: second frame memory 304: first table memory

305: block buffer 306: second table memory

307: adder 308: comparator

An apparatus of the present invention for achieving the above object comprises: a first block means for storing a table for quantizing a table lookup vector of subblocks in a block of a previous frame and a current frame; Second storage means for storing a vector quantization index of the previously decoded frame according to a previously decoded frame and the output of the first storage means; Third storage means for storing a current frame to be encoded; Fourth storage means for storing a vector quantization index of a block of the current frame and subblocks within the block according to the output of the first and third storage means; Address generating means for generating and specifying an address for reading data of the first storage means or writing data to the second to fourth storage means in accordance with the output of the second and fourth storage means; The difference value between subblocks precomputed to read the difference between the subblock in the block of the previous frame and the subblock in the block of the current frame according to the output of the second to fourth storage means. Fifth storage means for storing the data; Computing means for adding a difference between subblocks in blocks of the previous and current frames to obtain a difference between blocks of the previous and current frames according to the output of the fifth storage means; And comparing means for obtaining the spatial position difference by comparing the calculated inter-block difference value and the reference value from the computing means.

The present invention provides a block matched motion estimation method applied to a block matched motion estimation apparatus, comprising: dividing a block of a previous frame and a block of a current frame to be motion estimated into a plurality of subblocks in a motion estimation search region; Stage 1; A second step of storing a vector quantization index by quantizing a table lookup vector of each subblock in the block of the previous frame in the divided search region; A third step of storing a vector quantization index by quantizing a sub-blocks in a block of the current frame by a table lookup vector; A fourth step of reading difference values between subblocks of a subblock in the block of the previous frame and a subblock in the block of the current frame and summing the difference values to calculate an interblock difference; And a fifth step of finding the smallest inter-block difference among the calculated differences between the blocks of the current frame and the previous frame and transmitting the corresponding spatial position difference.

The present invention introduces a table lookup vector quantization technique to a macroblock difference calculation part of a motion estimator to increase the speed of the motion estimator.

To this end, in the present invention, the speed of the motion estimator can be improved by precomputing the difference between the representative subblocks and storing it in a memory and using it for motion estimation by replacing a large amount of computation required by the motion estimator with a table lookup.

The above objects, features and advantages will become more apparent from the following detailed description taken in conjunction with the accompanying drawings. Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

2 is a general block matched motion estimation partition diagram.

As shown in the figure, the block to estimate the motion of the current frame is commonly used in the difference calculation unit between a plurality of two blocks, there is no need to duplicate vector quantization in the difference calculation unit between a plurality of blocks, The quantization indices may be used by the vector quantization only in one difference calculation unit between two blocks of the two block differences.

Since the motion estimation search areas of the block to be estimated for current motion are overlapped, all subblocks of integer positions in the previous frame are vector quantized in advance, and the quantization indices are stored in a memory, and the corresponding motion estimation search areas are stored in the memory. Only quantization indices can be read from memory and used. Of course, when the amount of available memory is not sufficient, only quantization indexes corresponding to a part of the previous image frame may be stored, or a new quantization index may be obtained and used for each motion estimation block.

3 is a block diagram of an apparatus for estimating block matched motion using table lookup vector quantization according to an embodiment of the present invention.

As shown in the drawing, a block matched motion estimation apparatus using table lookup vector quantization according to the present invention includes an address generator for generating and designating a memory address for reading data in each memory or writing data to each memory. 301, a first frame memory 302 for storing a vector quantization index of a previous decoded frame and a previous decoded frame, a second frame memory 303 for storing a current frame to be encoded, a previous frame and a current frame block A first table memory 304 for storing the tables necessary to vector quantize each subblock of s, a block buffer 305 for storing the block of the current frame to be motion estimated and the quantization index of the subblocks in the block, vector Precalculates the difference between the quantized previous frame and the subblock of the current frame and stores the calculated value Second table memory 306 for calculating, an adder 307 for adding the difference between subblocks for calculating the difference between blocks, and a candidate having the smallest block difference by comparing the current block with a candidate block in the motion search region. Comparator 308 for finding the block.

As shown in the figure, the first frame memory 302 reads and stores the decoded previous frame, and the address generator 301 designates the address 1 to store the pixel value of the previous frame stored in the first frame memory 302. Read (data 1)

Subsequently, the first table memory 304 assigns the pixel value of the previous frame read by the address generator 301 as address 2 to obtain a vector quantization index obtained by quantizing a table lookup vector of all subblocks of integer positions in the previous frame. The frame memory 302 is stored.

Next, the second frame memory 303 reads and stores the current frame to be encoded, and the block buffer 305 uses the address 3 specified by the address generator 301 to store the current frame from the second frame memory 303. Read and store the block (data 3) to which the motion is to be estimated.

The address generator 301 then assigns address 4 to the block buffer 305 to read the pixel value (data 4) in the block of the current frame, and assigns this value to the first table memory 304 as address 2 to present it. A vector quantization index (data 5) in which a table lookup vector quantizes all subblocks in a block of a frame is stored in the block buffer 305.

Thereafter, the second table memory 306 uses the vector quantization index (data 6) of the subblock in the block of the previous frame and the vector quantization index (data 7) of the subblock in the block of the current frame to determine the difference between the subblocks. Read it.

Then, the adder 307 calculates the interblock difference by adding the difference between the subblocks (data 8) read out from the second table memory 306, and the comparator 308 calculates the interblock difference calculated by the adder 307 ( By comparing whether the data 9 is the minimum block difference, the spatial location difference (data 10) of the candidate block having the smallest difference and the current block is found. The spatial position difference found is the motion vector of the current block, which is transmitted to the decoder.

4 is a diagram illustrating an example of calculating a difference between two blocks according to the present invention, and relates to a case where a block size is 8x8 and a subblock is 4x2.

The motion estimation method firstly divides an arbitrary block (v in FIG. 4) and a current block (u in FIG. 4) to be estimated into several subblocks (a in FIG. 4). ), A function B for vector quantizing each subblocks of an arbitrary block (v in FIG. 4) and a function C for vector quantizing subblocks of the current block (u in FIG. 4) in the divided search region. C) of FIG. 4 and D function (d of FIG. 4) for reading the difference between sub-block representative values pre-computed and stored in memory after receiving quantization indexes from B and C functions. E function (eg, e of FIG. 4) that adds the difference values for the blocks.

As shown in the figure, in order to calculate the difference between the two blocks, each input block is first decomposed into two blocks, and a block v of the current frame for which the motion is to be estimated is also determined. Decompose in the same way. The decomposed subblocks are vector quantized through a table lookup vector quantizer, and the table lookup vector quantizer is composed entirely of table lookups without calculation. The current block u and the block v at any position in the search region may be vector quantized using different tables. As shown in FIG. 6, when the subblock is composed of eight pixels, seven table lookups (ie, 7/8 table lookups per pixel) are required to vector quantize one subblock. In general, when the size of the subblock is M × N, an average of 1-1 / M × N table lookups is required per pixel. The design method of the table lookup layer vector quantum is described in detail in several published papers.

Once each block is vector quantized, the vector quantization indices of the two subblocks at the same location in the two input blocks are used to approximate the SAD between the two blocks. For example, a unit block at the top in the leftmost current block u as _'11, the previous block within the unit block of the same location, v' _11, and the index value of the quantization by the self each block the vector quantity, each i, j Assuming that, the two blocks u _'11 and v' between SAD ₁₁ is approximated with (i, j) th value of t (i, j) in the place previously stored table. t (i, j) is a value stored in advance in the SAD between the representative vector x _i corresponding to the quantization index _i and the representative vector x _j corresponding to the quantization index j and stored in the table. Representative vectors x _j and x _i do not need to be stored in memory, and the method of obtaining these representative vectors may use any of several vector quantizer design methods already published in many papers.

By adding all the SADs between the approximated subblocks, the SAD between the two input blocks can be approximately calculated. As shown in FIG. 5, when one block is composed of eight subblocks, eight table lookups and seven additions are required for this purpose. In general, when the block size is L × L and the subblock size is M × N, L / M × L / N table lookups and L / M × L / N−1 addition operations are required. As described above, the proposed motion estimation method leads to a significant reduction in computation.

For example, if the size of a block in motion estimation unit is 16 × 16, the size of a subblock in quantization unit is 4 × 2, and the motion estimation search region is + -16 horizontally and + -16 vertically, one pixel. An average of 140 table lookups, 132 additions, and 4 comparison operations are required per session. This is a very small amount of computation compared to the conventional method (1089 addition and subtraction per pixel and 1093 comparison operations).

The present invention can be applied to values other than the parameters of the above-described embodiments. That is, the size of the block can be applied to 8x8, 16x16, or other sizes, and the size of the subblock can also be used other than 4x2. Increasing the size of the subblock increases the speed by decreasing the amount of computation required instead of decreasing the accuracy of the motion estimation, and decreasing the size of the subblock increases the speed increase for the conventional method instead of increasing the accuracy of the motion estimation. In addition to the SAD, an SSD or other block difference measure may be used.

The present invention can be applied to various existing fast motion estimation methods such as full-search motion estimation method as well as three step motion estimation method. The principle of the conventional fast motion estimation method, such as the global search motion estimation method and the three-step motion estimation method, can be easily found in various published papers.

FIG. 5 is a diagram illustrating a relationship between a block, which is a motion estimation unit, and a subblock, which is a vector quantization unit, according to an embodiment of the present invention. Indicates.

FIG. 6 is an exemplary explanatory diagram illustrating hierarchical vector quantization of each subblock according to the present invention, and illustrates a case where the size of the subblock to be quantized vector is 4 × 2.

7 is a flowchart illustrating a block matched motion estimation method using table lookup vector quantization according to the present invention.

As shown in the figure, the block matched motion estimation method using the table lookup vector quantization according to the present invention first reads the pixel value of the decoded previous frame (701), and converts the subblock of the integer position of the previous frame into a table lookup vector. Quantization stores a vector quantization index (702), reads a block of the current frame (703), and stores a vector quantization index by quantizing a table lookup vector of subblocks in the block of the current frame (704).

Then, the vector quantization index of the subblocks in the candidate block is read (705), the subblock difference values are read from the vector quantization index of the current block and the subblocks in the candidate block (706), and the block difference is added by adding the difference values between the subblocks. The value is calculated (707).

Next, it is determined whether the calculated block difference value is the minimum block difference (708). If the minimum block difference is determined, the spatial difference between the candidate block and the current block is stored (709). If there are still more candidate blocks (710), if there are still more candidate blocks, the method reads the vector quantization indexes of the subblocks in the next candidate block (711), and reads the difference between the subblocks from the vector quantization indexes in the current block and the candidate blocks. Repeat it.

On the other hand, if it is not the minimum block difference, it is determined whether more candidate blocks remain in the search area (710), if there are no more candidate blocks, the spatial position difference is transmitted (712), and whether more blocks remain in the current frame If there are no more blocks in the current frame (713), the next frame is processed. If there are more blocks in the current frame, the next block of the current frame is read (714). The process of storing the vector quantization index is repeated.

As mentioned above, the method proposed in the present invention only requires much less computation amount for motion estimation as compared to the prior art. For example, when the size of a block, which is a motion estimation unit, is 16 × 16, the size of a subblock, which is a quantization unit, is 4 × 2, and the moving search region is + -16 pixels horizontally and + -16 pixels vertically, The method requires approximately 1089 additions and subtractions and 1093 comparison operations per pixel. However, in the present invention, only 140 table lookups, 132 additions, and 4 comparisons are required per pixel. Given that much of the video codec's current computation comes from the motion estimator, this reduction in the amount of computation enables real-time software video codec implementation even on low-performance computers. Suitable for

In the present invention, since the difference between sub-blocks is calculated in advance and stored in a table, it is possible to use a complicated difference measure that is difficult to calculate in real time. Therefore, it is easy to use a new standard for motion estimation that can improve video compression performance.

In the present invention, a trade-off between the amount of memory and the motion estimation performance and the trade-off between the motion speed and the estimation performance are possible, so that the motion estimator can be adaptively implemented according to the usage environment such as a desktop computer, a notebook computer, and a PDA. It is possible.

The present invention described above is capable of various substitutions, modifications, and changes without departing from the spirit of the present invention for those skilled in the art to which the present invention pertains, and the above-described embodiments and accompanying It is not limited to the drawing.

Therefore, the present invention as described above has the effect of reducing the amount of calculation required for motion estimation by reading values previously stored in the memory, instead of actually calculating the difference between blocks for motion estimation.

In addition, the present invention divides each input block into several subblocks and quantizes the divided subblocks through a table lookup vector quantizer, and then stores the difference between the quantized blocks previously calculated using the quantization indices and stored in the memory. It is possible to improve the speed of the motion estimator by replacing the large amount of computation required by the motion estimator with a table lookup.

Claims (4)

In the block-matched motion estimation apparatus, First storage means for storing a table for quantizing a table lookup vector of subblocks in a block of a previous frame and a current frame; Second storage means for storing a vector quantization index of the previously decoded frame according to a previously decoded frame and the output of the first storage means; Third storage means for storing a current frame to be encoded; Fourth storage means for storing a vector quantization index of a block of the current frame and subblocks within the block according to the output of the first and third storage means; Address generating means for generating and specifying an address for reading data of the first storage means or writing data to the second to fourth storage means in accordance with the output of the second and fourth storage means; The difference value between subblocks precomputed to read the difference between the subblock in the block of the previous frame and the subblock in the block of the current frame according to the output of the second to fourth storage means. Fifth storage means for storing the data; Computing means for adding a difference between subblocks in blocks of the previous and current frames to obtain a difference between blocks of the previous and current frames according to the output of the fifth storage means; And Comparison means for obtaining a spatial position difference by comparing the calculated inter-block difference value from said computing means with a reference value Block matching motion estimation apparatus using table lookup vector quantization comprising a. In the block matching motion estimation method applied to the block matching motion estimation apparatus, Dividing a block of a previous frame and a block of a current frame to be motion estimated into a plurality of subblocks in a motion estimation search region; A second step of storing a vector quantization index by quantizing a table lookup vector of each subblock in the block of the previous frame in the divided search region; A third step of storing a vector quantization index by quantizing a sub-blocks in a block of the current frame by a table lookup vector; A fourth step of reading difference values between subblocks of a subblock in the block of the previous frame and a subblock in the block of the current frame and summing the difference values to calculate an interblock difference; And A fifth step of finding the smallest block difference among the calculated blocks of the current frame and the previous frame and transmitting the corresponding spatial position difference Block matching motion estimation method using a table lookup vector quantization comprising a. The method of claim 2, The fourth step, A sixth step of reading difference values between subblocks from a vector quantization index of subblocks in the current block and the candidate block; And A seventh step of calculating the difference between blocks by adding the difference between subblocks Block matching motion estimation method using a table lookup vector quantization comprising a. The method of claim 3, wherein The fifth step, An eighth step of determining whether the calculated difference value between blocks is a minimum block difference; A ninth step of determining whether the candidate block remains in the search area by storing the spatial position difference between the candidate block and the current block if the minimum block difference is determined as the result of the eighth step; A tenth step of reading quantization indices of subblocks in a next candidate block if the candidate block remains, and repeating the operation from the sixth step; An eleventh step of transmitting a spatial position difference and checking whether more blocks remain in the current frame if there are no more candidate blocks as a result of the determination of the ninth step; A twelfth step of reading the next block of the current frame if more blocks remain, repeating from the storing step of the vector quantization index of the fifth step, and reading the next frame if no more blocks remain; And As a result of the determination of the eighth step, if it is not the minimum block difference, the thirteenth step is performed from the process of determining whether the candidate block of the ninth step remains Block matching motion estimation method using a table lookup vector quantization comprising a.