KR100747544B1 - Motion estimation method and apparatus - Google Patents

Abstract

A motion estimation method and an apparatus therefor are provided to quicken a motion estimation speed and improve coding quality by applying an adaptive threshold value and correcting the number of candidate MVs(Motion Vectors) for selecting the optimum MV of a current block to be encoded. An integer MV search unit(10) searches an MV of an integer pixel unit for a current block by using matching measurement values between the current block to be encoded and respective reference blocks of a search region in a reference frame. The first MV candidate selecting and storing units(20,30) compare the matching measurement values of the integer pixel unit with the first threshold value, and select and store the first MV candidates from MVs of the integer pixel unit. A sub-pixel MV search unit(40) searches an MV of a sub-pixel unit for the current block in a local region where all the first MV candidates are located. The second MV candidate selecting and storing units(50,60) compare matching measurement values of the sub-pixel unit with the second threshold value, and select and store the second MV candidates from MVs of the sub-pixel unit. A rate distortion MV search unit(70) analyzes a bit rate distortion on the basis of the second MV candidates, and selects the optimum predictive mode for the current block.

Description

Motion estimation method and apparatus

1A and 1B show examples of a general sub-pixel motion estimation method.

2 is a graph showing the relationship of PSNR vs. bit rate for a test video sequence that depends on the number N of motion vector candidates.

3 shows an example in which the PSNR for the same video sequence under a fixed bit rate depends on the candidate number N of motion vectors.

Figure 4 is a block diagram showing an embodiment of a motion estimation apparatus according to the present invention

Explanation of symbols for main parts of the drawings

10: integer MV search unit 20: first MV candidate selection unit

30: first MV candidate storage 40: sub-pixel MV searcher

50: second MV candidate selecting unit 60: second MV candidate storing unit

70: rate distortion MV search unit

The present invention relates to digital image compression, and more particularly to a method and apparatus for block matching motion estimation.

In general, because of the very large size of the uncompressed digital data, compression is applied to the video signals, and as a result can be transmitted and stored. That is, various types of image compression algorithms for digitally encoded image data have been developed to minimize the bandwidth required for transmitting digitally encoded image data (hereinafter, digital image data).

Typically ISO Moving Pictures Expert Group (MPEG-1), MPEG-2, MPEG-4, Telecommunication Standardization Sector of the International Telecommunication Union (ITU-T) H.261, H.263, H.263 +, H.264 / There are many international standards, including AVC. And they are commonly used for video distribution such as VCD, DVD, DVB, HDTV, video conferencing, video editing, video streams on the Internet, and the like.

A general feature of one of these standards is that motion estimation is used to reduce inherent temporal redundancy in video sequences. In motion estimation, each frame is typically divided into square blocks or macroblocks containing 16x16 pixels. For each of those current blocks / macroblocks, a full exhaustive search is typically performed within the search area of the predefined reference frame. The reference frame may be a previous frame or a future frame most similar to a block / macroblock to be currently encoded according to a predefined criterion function in one video sequence. In the modern coding standard, motion estimation for each frame or field is performed based on a plurality of reference frames.

The matching measurement function is typically a sum of the absolute difference (SAD). Another common match measurement function includes a sum of the square difference (SSD). That is, a matching measurement function (eg, SAD or SSD) between each current block / macroblock and each reference block / macroblock in the search region of the reference frame is obtained, and the reference block / macroblock of the smallest measurement function is It becomes a reference block / macro block most similar to the current block / macroblock. As described above, motion estimation is a process of obtaining a motion vector indicating a position difference of a reference block / macroblock most similar to the current block / macroblock using a predetermined measurement function. Here, the SAD value means the absolute value of the difference between each reference block and the current block in the reference frame.

Thus, the current block may be compression coded as the difference between the motion vector, the most similar block corresponding to the motion vector of the current block and the reference frame.

In modern coding standards, motion vectors are searched separately for all divisions and subdivisions of macroblocks, for example 16x16, 8x16, 16x8, 8x8, 4x8, 8x4, and 4x4 subblocks. .

As described above, motion vectors are searched for every frame or all pictures from a plurality of reference frames. In addition, the search for motion vectors is performed with subpixel accuracy (half-pixel or quarter-pixel accuracy) to obtain good coding quality. Here, motion estimation is a very important and time consuming part of the video encoding process.

However, it is well known that the exhaustive full search is very slow.

Therefore, it is highly desirable to develop a fast motion estimation algorithm that does not adversely affect the image quality of an image that can be reproduced from a compressed image signal.

1A and 1B illustrate exemplary embodiments of a conventional motion estimation method.

Referring to FIG. 1A, it consists of a preliminary full interpolation of the current and reference frame and a search in the interpolated search space.

1B is characterized by a hierarchical search. That is, a global integer motion vector search is performed to find an optimal integer motion vector having a minimum measurement function. Next, a sub-pixel motion vector search is performed in a local neighborhood found as the optimal motion vector to obtain a motion vector of the current block.

The advantages and disadvantages of these motion estimation methods are obvious. Figure 1A achieves good motion estimation results, but has the disadvantage of being very labor-intensive and slow. Although FIG. 1B is very fast, the global extremum cannot be calculated, which is a disadvantage in terms of quality.

Numerous experiences with test video sequences are shown in the following facts. If after the first search step, we select N motion vector candidates instead of the most optimal motion vectors and perform a local sub-pixel search for each of them, the coding quality will be essentially higher.

FIG. 2 is a graph showing a relationship between a peak signal to noise ratio (PSNR) versus a bit rate of a test video sequence 'FOREMAN.QCIF' for N motion vector candidates having different values.

Referring to FIG. 2, it can be seen that coding quality increases as the number N of motion vector candidates increases.

3 shows an example in which the average PSNR varies depending on the number of motion vector candidates at a fixed bit rate (ie, 75 Kbit / sec) for the same video sequence.

Referring to FIG. 3, when the typical size of the search region is 1089 (= (2 * search_range + 1) * (2 * search_range + 1), where search_range = 32), the graph shows that the number of motion vector candidates is 50. It seems to be most satisfying when it is possible, and the result shows that further increase is meaningless.

Accordingly, an object of the present invention is to provide a method and apparatus for quickly and simplifying a motion estimation process by performing adaptive motion estimation with a hierarchical motion vector search process.

Motion estimation method according to the present invention for achieving the above object,

(a) searching for a motion vector of an integer pixel unit for the current block by using a matching measure between a current block to be encoded and each reference block of a search region in a reference frame;

(b) selecting first motion vector candidates from the motion vectors of the integer pixel unit found in the step (a) by comparing the matching measurement value of the integer pixel unit and a first threshold value;

(c) searching for a motion vector on a sub-pixel basis for a current block in a local region where all of the first motion vector candidates are collected;

(d) selecting second motion vector candidates from the motion vectors of the sub-pixel units found in the step (c) by comparing the matching measurement value of the sub-pixel units and a second threshold value; And

and (e) analyzing the bit rate distortion based on the second motion vector candidates to select an optimal prediction mode for the current block.

The matching measurement value is calculated as a sum of absolute difference (SAD) between the current block and the motion compensated block.

The matching measurement value is calculated as a sum of square difference (SSD) between the current block and the motion compensated block.

The step (b) may further include storing the selected first motion vector candidates and recalculating the first threshold.

The step (d) may further include storing the selected second motion vector candidates and recalculating the second threshold.

According to an aspect of the present invention, there is provided a motion estimation apparatus comprising: an integer MV search unit for searching for a motion vector of an integer pixel unit for the current block by using a matching measurement value of a current block to be encoded and each reference block of a search region in a reference frame; A first MV candidate selection and storage unit comparing the matching measurement value in the integer pixel unit with a first threshold value to select and store first motion vector candidates from the motion vectors in the integer pixel unit; A sub-pixel MV search unit for searching for a motion vector of a sub-pixel unit for the current block in a local region where all of the first motion vector candidates are collected; A second MV candidate selection and storage unit comparing the matching measurement value in the sub-pixel unit with a second threshold value to select and store second motion vector candidates from the motion vectors in the sub-pixel unit; And a rate distortion MV searcher for selecting an optimal prediction mode for the current block by analyzing bit rate distortion based on the second motion vector candidates.

Other objects, features and advantages of the present invention will become apparent from the following detailed description of embodiments taken in conjunction with the accompanying drawings.

Hereinafter, with reference to the accompanying drawings illustrating the configuration and operation of the embodiment of the present invention, the configuration and operation of the present invention shown in the drawings and described by it will be described as at least one embodiment, By the technical spirit of the present invention described above and its core configuration and operation is not limited.

The present invention divides the current frame picture into a set of macroblocks and divides each of them into a set of blocks or subblocks of different sizes.

The present invention performs a first search using the original resolution of the blocks and macroblocks in the current frame and the reference frame to be encoded. Here, the reference frames are frames before or after the frame to be encoded on the video sequence. The search may be a full (or exhaustive) search or some kind of fast search technique. When the first search step is performed, the number of motion vector candidates is determined. These motion vector candidates are selected by comparing the defined matching measurement function, i.e. the calculated matching measurement value (e.g., SAD value), with two adaptive thresholds. The first of the two thresholds is equal to the most recently reached best match measure, whereas the second threshold is less than the first threshold on one displacement.

Here, the threshold value is estimated in consideration of the possibility to improve the matching measure in the next search step. All candidate motion vectors pass the second step of motion estimation. In a second step, the selection from the candidate motion vectors is performed using the subpixel (half-pixel and / or quarter-pixel) resolution of blocks and macroblocks in the current frame and the reference frame.

The sub-pixel search is performed in a local region where all of the motion vector candidates are concentrated, and the sub-pixel interpolation is required only for these local regions, not for the entire reference frame. The optimal motion vector is the result of the second phase search, taking into account the measurement of the bit rate / distortion analysis performed on all motion vectors found as a result of the second search step, as well as matching measurement functions such as SAD or SSD. Can be selected from.

4 shows a block structure of the most preferable motion block estimator of the present invention, wherein an integer MV search unit 10, a first MV candidate selector 20, a first MV candidate storage unit 30, and a sub-pixel MV are shown. The searcher 40 includes a searcher 40, a second MV candidate selector 50, a second MV candidate storage 60, and a rate distortion MV searcher 70.

In the present invention configured as described above, motion estimation for one block (its size may vary from a full macroblock to a smaller subblock) is an integer MV search unit 10 similar to the conventional motion vector search method of FIG. Starts at). However, the result of the search is not only the most optimal motion vector found according to the search measurement function but also the optimal motion vector selected by the first MV candidate selector 20 and stored in the first MV candidate storage 30. The full set of candidates.

In the integer MV search section 10, the scanning of the MV search area is performed in the helical order, and the matching measurement value Si is calculated for all reference blocks and the current block in the search area. The calculated matching measurement value Si becomes one of the inputs of the first MV candidate selector 20. For another input, the first MV candidate selector 20 receives the selection thresholds T1, T2 from the first MV candidate store 30.

The selection thresholds T1 and T2 have adaptive values as shown below.

That is, the selection threshold T1 is equal to the optimal matching measurement value for the current block obtained until the time when the entire search area is scanned, as shown in Equation 1 below.

The selection threshold T2 is larger than T1 by shift ΔT as shown in Equation 2 below.

The shift ΔT between the T1 and T2 is equal to an upper boundary that reduces all possible matching measurement functions during the sub-pixel search.

The shift value ΔT depends on various factors and may be converted during the entire search process, but at the beginning of the process it must be defined in advance.

When the first MV candidate selector 20 receives the matching measurement value Si and the selection thresholds T1 and T2, the following comparisons are performed. If the selection threshold T2 satisfies the condition of Equation 3 below, i.e., if the matching measurement Si is less than or equal to the selection threshold T2, the i-th reference block corresponding to the matching measurement Si is set of MV candidates. Will be included.

Further, if the selection threshold T1 satisfies the condition of Equation 4 below, that is, if the matching measurement value Si is less than or equal to the selection threshold T1, the selection threshold T1 is the matching measurement as shown in Equation 5 below. The value Si is reassigned, and the selection threshold T2 must be recalculated according to Equation 2 above.

T1 = Si

In this case, the first MV candidate selector 20 selects motion vectors that provide a matching measurement value Si that satisfies the condition of Equation 3, and applies the motion vectors to the first MV candidate storage 30.

As soon as the first MV candidate storage unit 30 receives the current motion vector from the first MV candidate selecting unit 20, the first MV candidate storage unit 30 stores the motion vector in the first MV candidate storage unit 30. Operations must be performed. If the condition of Equation 4 is satisfied, and then reassignment of the selection threshold T1 is performed as in Equation 5, it is necessary to correct the set of MV candidates and discard all the missing MV candidates. That is, MV candidates corresponding to the matching measurement value larger than the selection threshold before being reallocated are discarded.

And other embodiments of limiting the number of possible candidates, including ordinal storage and methods of using the set, may be other embodiments of the present invention.

In all cases, however, all MV candidates that no longer satisfy the condition of Equation 3 above are applied with the new T2 value and are no longer meaningful and will be excluded from the set of MV candidates.

The set of corrected first MV candidates output from the first MV candidate storage 30 is applied to the input of the sub-pixel MV searcher 40. In the sub-pixel MV search section 40, a motion vector search with sub-pixel precision is performed in the local vicinity of all MV candidates in the set.

Similar to what has been described previously, the result of the search is not only the optimal motion vector value found according to the sub-pixel search criteria, but also selected by the second MV candidate selector 50 to be sent to the second MV candidate storage 60. Total candidate set for the stored optimal motion vector value.

In the sub-pixel MV search section 40, scanning the local vicinity of all MV candidates in the set is performed with half-pixel precision in helical order and quarter precision in double-helical order. Similar to what has been described previously, the matching measure is calculated for the current block and the reference block at all locations in its vicinity. The calculated matching measure becomes one of the inputs of the second MV candidate selector 50. For another input, the second MV candidate selector 50 receives the selection thresholds from the second MV candidate storage 60. The selection thresholds mean that the optimized time level has been reached, which level may be reached in the next step. Here, the number of selected first motion vector candidates and the number of selected second motion vector candidates are limited by constants N1 and N2 (where N1> N2).

All operations of processing the MV candidates to select thresholds are similar to those described above. The set of corrected MV candidates output from the second MV candidate storage 60 is applied to the input of the rate distortion MV selector 70. In the rate distortion MV selector 70, the selection of the optimum mode for each macroblock is made through the intra mode. The type of intra prediction mode depends on the coding standard. For example, the H.264 / AVC standard provides seven inter modes and one skip mode. The selection is performed by an algorithm, and the algorithm calculates and evaluates in advance the bits for encoding the current macroblock and distortions that may occur when encoding the macroblock in the mode.

The mode selection is a critical, time consuming step, and only a few candidates can participate in its solution. After all, the optimal motion vector of the current block is determined only after the selection of the optimal prediction mode for the macroblock is performed.

As described above, according to the motion estimation method and apparatus according to the present invention, the motion estimation speed is increased by correcting the number of candidate motion vectors for selecting the optimal motion vector of the current block to be encoded by applying an adaptive threshold value. There is an effect of improving the encoding quality.

Those skilled in the art will appreciate that various changes and modifications can be made without departing from the spirit of the present invention.

Therefore, the technical scope of the present invention should not be limited to the contents described in the embodiments, but should be defined by the claims.

Claims (11)

(a) searching for a motion vector of an integer pixel unit for the current block by using a matching measure between a current block to be encoded and each reference block of a search region in a reference frame; (b) selecting first motion vector candidates from the motion vectors of the integer pixel unit found in the step (a) by comparing the matching measurement value of the integer pixel unit and a first threshold value; (c) searching for a motion vector on a sub-pixel basis for a current block in a local region where all of the first motion vector candidates are collected; (d) selecting second motion vector candidates from the motion vectors of the sub-pixel units found in the step (c) by comparing the matching measurement value of the sub-pixel units and a second threshold value; And (e) analyzing a bit rate distortion based on the second motion vector candidates to select an optimal prediction mode for the current block. The method of claim 1, wherein the matching measure is And calculating a sum of absolute difference (SAD) between the current block and the motion compensated block. The method of claim 1, wherein the matching measure is And calculating the sum of square difference (SSD) between the current block and the motion compensated block. The method of claim 1, wherein step (b) Storing the selected first motion vector candidates, and recalculating the first threshold. The method of claim 1, wherein step (d) Storing the selected second motion vector candidates and recalculating the second threshold. The method according to any one of claims 4 to 5, The recalculation of the first and second thresholds compares the matching measurements with the most recently entered optimal matching measure and discards all motion vector candidates that are not desired in a subsequent search step. Motion estimation method, characterized in that by correcting. An integer MV search unit for searching for a motion vector of an integer pixel unit for the current block by using a matching measurement value of each current block to be encoded and each reference block of the search region in the reference frame; A first MV candidate selection and storage unit comparing the matching measurement value in the integer pixel unit with a first threshold value to select and store first motion vector candidates from the motion vectors in the integer pixel unit; A sub-pixel MV search unit for searching for a motion vector of a sub-pixel unit for the current block in a local region where all of the first motion vector candidates are collected; A second MV candidate selection and storage unit comparing the matching measurement value in the sub-pixel unit with a second threshold value to select and store second motion vector candidates from the motion vectors in the sub-pixel unit; And And a rate distortion MV searcher for analyzing a bit rate distortion based on the second motion vector candidates to select an optimal prediction mode for the current block. 8. The method of claim 7, wherein the matching measure is And a sum of absolute difference (SAD) between the current block and the motion compensated block. 8. The method of claim 7, wherein the matching measure is And a sum of square difference (SSD) between the current block and the motion compensated block. The method of claim 7, wherein the first MV candidate selection and storage unit And if the matching measurement value is less than or equal to a first threshold value, updating the first threshold value to a matching measurement value. 11. The method of claim 10, wherein the second MV candidate selection and storage unit And updating the second threshold value by adding a constant? T to the first threshold value when the first threshold value is updated.

Images