CN104702957B - Motion vector compression method and device - Google Patents

Abstract

The invention discloses a method and device for compressing motion vectors to solve the problems of long time-consuming compression and encoding of motion vectors and limited performance in the prior art. The method includes searching the pixel precision area of the motion vector, searching the motion vector according to two different pixel precision areas; expressing the MV of the first area by using the first expression, and using the second expression for the MV of the second area Two notation for representation. In the process of motion search, the present invention performs partition search on the motion vector according to different pixel precision regions, discards some pixels in the low pixel precision region, and then properly converts the MV of different pixel precision regions and uses Different representation methods are used to express, so as to achieve the purpose of compressing the motion vector, encoding the compressed motion vector, reducing the code stream, optimizing the performance of motion estimation, improving the speed of motion estimation, and realizing the gain of coding performance.

Description

Motion vector compression method and device

technical field

The invention belongs to the technical field of data compression, and in particular relates to a motion vector compression method and device.

Background technique

In the field of video compression technology, the international mainstream coding standard is H.264/AVC, the first generation video compression coding standard in my country is MPEG-2, and AVS is the second generation source coding standard with independent intellectual property rights in my country. In 2012, AVS+ introduced advanced field technology and entropy coding technology, and officially became the broadcasting industry standard. At present, my country's upcoming AVS2 standard is the next-generation standard of AVS.

AVS2's quadtree-based block mode and flexible reference frames greatly improve the coding performance of AVS2 for high-definition video sequences, thus saving up to 50% of the bandwidth compared to the previous international mainstream coding standard H.264/AVC Above, the subjective quality has also been greatly improved, and has a broader application space. For example, for the block size, the largest coding unit (CTU, Coding Tree Unit) has been increased from 16x16 of AVS to 64x64 of AVS2, and the coding unit can be divided from 64x64 to 8x8, so that the coding can adapt to more variable resolutions and more Complex image textures.

However, the complexity of the entire encoder of AVS2 increases several times with the improvement of performance, and inter-frame encoding occupies the main time of encoding (except for the encoding configuration of all I frames). The most time-consuming part of inter-coding is motion estimation, which comes from the process of searching for matching blocks at integer and sub-pixel precision. The compression encoding of the motion vector in the existing AVS2 data compression process still has the problem of taking too long. In addition, the compression performance of the final motion vector in the inter-frame prediction process is the main factor affecting the compression performance of AVS2. Because there are certain limitations in the compression performance of the motion vector in the prior art, the accuracy of the inter-frame prediction is not high, thus There is still a gap between the AVS2 standard and the latest video coding standard HEVC jointly released by the ITU-T VCEG and the ISO/IEC MPEG standardization organization.

Contents of the invention

The purpose of the present invention is to provide a motion vector compression method and device, by introducing progressive motion vector accuracy, to optimize the motion estimation performance, improve the motion estimation speed, and realize the gain of coding performance.

According to one aspect of the present invention, a method for compressing motion vectors is provided, the method comprising:

searching the pixel precision region of the motion vector, dividing the motion vector into two regions, and searching the two regions using different pixel precision;

The MV of the first region is represented by the first notation, and the MV of the second region is represented by the second notation.

In the above solution, the two different pixel precision regions are further a first region with 1/4 pixel precision and a second region with 1/2 pixel precision.

In the above solution, when the motion vector is located in the second area with 1/2 pixel precision and itself has 1/4 pixel precision, the pixel corresponding to the motion vector is discarded.

In the above solution, the search for the pixel-accurate region of the motion vector is realized by a threshold.

In the above solution, the MV of the second region is represented by the second representation, further, the MV is compressed at the encoding end, and the MV is restored at the decoding end.

In the above scheme, the MV is compressed at the encoding end, and the MV is further converted using the Algorithm 1 formula, and the MV is restored at the decoding end, and the MV is further converted using the Algorithm 2 formula The above MV is converted.

According to another aspect of the present invention, a motion vector compression device is also provided, the device comprising:

a search unit, configured to search for a pixel precision region of the motion vector, divide the motion vector into two regions, and search the two regions using different pixel precision;

A conversion unit, which is connected to the search unit and configured to express the MV of the first area by using the first notation, and express the MV of the second area by using the second notation.

In the solution above, the search unit is further configured to search the motion vector according to different areas of the first area with 1/4 pixel precision and the second area with 1/2 pixel precision.

In the above solution, the search unit is further configured to discard the motion vector corresponding to the motion vector when the searched motion vector is located in the second area of 1/2 pixel precision and the motion vector itself is 1/4 pixel precision. of pixels.

In the above solution, the conversion unit expresses the MV of the second area using the second representation, and further comprises: compressing the MV at the encoding end, and recovering the MV at the decoding end.

It can be seen from the above technical solutions of the embodiments of the present invention that the motion vector compression method of the present invention performs partition search on the motion vector according to different pixel precision regions during the motion search process, and the low pixel precision region Part of the pixels are discarded, and then the MVs in different pixel precision areas are properly converted and represented by different representation methods, so as to achieve the purpose of compressing the motion vector, encoding the compressed motion vector, reducing the code stream, optimizing The performance of motion estimation is improved, the speed of motion estimation is improved, and the gain of coding performance is realized.

Description of drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings that need to be used in the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For Those of ordinary skill in the art can also obtain other drawings based on these drawings without any creative effort.

FIG. 1 is a schematic flow chart of a motion vector compression method according to a first embodiment of the present invention;

FIG. 2 is a schematic flowchart of a motion vector compression method according to a second embodiment of the present invention;

FIG. 3 is a schematic diagram before motion vector compression in the second embodiment of the present invention;

4 is a schematic diagram of compressed motion vectors in the second embodiment of the present invention;

FIG. 5 is a schematic structural diagram of a motion vector compression device according to a third embodiment of the present invention.

Detailed ways

Embodiments of the present invention are described in detail below, examples of which are shown in the drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the figures are exemplary only for explaining the present invention and should not be construed as limiting the present invention.

Those skilled in the art can understand that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. It should also be understood that terms such as those defined in commonly used dictionaries should be understood to have a meaning consistent with the meaning in the context of the prior art, and will not be interpreted in an idealized or overly formal sense unless defined as herein explain.

The technical problem to be solved by the present invention is to improve the compression rate and compressibility of the final motion vector in the inter-frame prediction process. In the AVS2 standard, there are dual problems of motion estimation complexity and too many motion vector coding bits in the inter-frame prediction process. Therefore, there is still a lot of room for optimization of motion estimation performance and improvement of motion speed. In the actual compression coding process, statistics and analysis of the inter-frame prediction behavior of the final motion vector, the distribution of the motion vector is obtained, showing the trend of motion vectors gathered near the MVP, which makes the use of motion vector statistical information to optimize motion Vectors became possible. On the basis of the above-mentioned statistical results, the present invention proposes the concept of asymptotic motion vector accuracy and introduces it into the compression of motion vectors, according to motion vector (MV, Motion Vector) and motion vector prediction value (MVP, Motion The difference in the relative position of VectorPredictor) imposes different restrictions on the accuracy of the MV search, thereby adjusting the search process of the motion vector, and at the same time improving the compression effect of the MV amplitude, thereby reducing the encoding bit rate.

At present, in the AVS standard, MVs are fixedly represented by 1/4 pixel precision, and the present invention uses higher precision for MVs relatively close to the MVP, and uses lower precision for MVs relatively far away from the MVP, so that the closer to the MVP The MV of is more likely to obtain the optimal rate-distortion cost. When specifically implemented in the encoder, use high pixel precision for MVs within a certain range that are close to the MVP, such as 1/4 pixel precision, and use low pixel precision for MVs that are far away from the MVP within a certain range, such as 1/4 2 pixel precision. The high and low here are relative concepts relative to approach and distance. In actual implementation, an infinitely small range can be distinguished to realize the distance between MV and MVP from near to far. Relatively used to represent MV The asymptotic process of pixel precision from high to low, that is, progressive motion vector precision (PMVR, Progressive Motion Vector Resolution).

The technical scheme of the present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.

FIG. 1 is a schematic flowchart of a motion vector compression method according to a first embodiment of the present invention.

As shown in Figure 1, the motion vector compression method of the present embodiment includes the following steps:

Step S1, searching the pixel precision area of the motion vector, dividing the motion vector into two areas, the two areas include a first area and a second area, and searching the two areas using different pixel accuracy .

In this step, the motion vector is divided into two regions, which can be realized by a threshold (TH). In the actual implementation process, when the motion vector is limited to two areas, it can be 1/4 pixel precision and 1/2 pixel precision, such as [CTR-TH, CTR+TH] area with 1/ Search for motion vectors with 4-pixel precision, and search for motion vectors with 1/2-pixel precision outside the area. Wherein, the CTR represents the starting point of the vector representation.

On the basis of the above-mentioned area limitation, when performing motion search, the stepwise refinement method in the prior art can be adjusted, such as setting the first area to be of higher precision, and the second area to be of lower precision, when the candidate When the MV of MV exceeds the accuracy range of the first area, the MV is ignored and does not need to be searched, thereby eliminating part of the MV search of the accuracy of the first area, and reducing the complexity of motion estimation. In the specific implementation process, it is shown to judge the MV. If |MVx-CTRx|>TH or |MVy–CTRy|>TH, and the MV is 1/4 pixel precision, then this point is discarded and does not participate in the best MV The decision-making reduces the complexity of motion estimation.

In step S2, the MV of the first area is represented by the first notation, and the MV of the second area is represented by the second notation.

The expression method here is to use different representation methods for the regions with different precisions on the basis of restricting the regions with different precisions for the motion vector, so as to reduce the number of encoded bits without losing the information amount of the MV. For example, for MV in the first area with 1/4 pixel precision, use 1/4 precision to express; and for MVD or MV in the second area with 1/2 pixel precision, use Algorithm 1 formula or Algorithm 2 formula to convert. Wherein, the Algorithm 1 formula conversion of the MVD corresponds to the encoding end, and the Algorithm 2 formula conversion of the MV corresponds to the decoding end.

It should be noted that the present invention is applicable to motion vectors in encoders of any encoding standard.

The motion vector compression method of this embodiment reduces the complexity of the encoding end by dividing MVs with different distances into different precision areas, and realizes the limitation or segmentation of the above-mentioned areas through methods such as thresholds, and then uses different representation methods to optimize The performance of motion estimation improves the speed of motion estimation, and at the same time improves the compression effect of MV amplitude, thereby reducing the encoding bit rate, thereby achieving the gain of encoding performance.

FIG. 2 is a schematic flowchart of a motion vector compression method according to a second embodiment of the present invention.

As shown in FIG. 2 , this embodiment takes a motion vector as an example, and uses the PMVR method to compress the motion vector. The motion vector here, the motion vector can be any standard.

The compression method of described motion vector, comprises the steps:

Step S21, searching the motion vector to obtain the first MV region with 1/4 pixel precision and the second MV region with 1/2 pixel precision.

In this step, the PMVR method indicates the use area of 1/4MV precision by setting a threshold TH. In this embodiment, the value of TH is 2, as shown in FIG. 3 and FIG. 4 . TH is a threshold that controls the available range of 1/4 pixel precision (in units of 1/4 pixel, TH is a non-negative integer and the value is an integer multiple of 2), and the area in the red box is the threshold TH Within the limited range, that is, the area enclosed by the red square, that is, the first area, the MV is searched with a precision of 1/4 pixel. MVs outside the first area can only be searched with 1/2 pixel accuracy, and the area they belong to is the second area. CTR (CenTer of the Range) is the center of the first area.

As shown in Figure 3, when the MVP points to a half-pixel or full-pixel precision position, the CTR and MVP overlap; as shown in Figure 4, when the MVP points to a 1/4 pixel position, the CTR and MVP each occupy a pixel. position, not overlapping.

When the CTR does not overlap with the MVP, the CTR is obtained by rounding the MVP through formula (1)-(2), and the formula (1)-(2) is as follows:

CTRx=MVPx>>1<<1 (1)

CTRy=MVPy>>1<<1 (2)

Wherein, the CTRx indicates the component of the CTR in the X direction, MVPx indicates the component of the MVP in the X direction; the CTRy indicates the component of the CTR in the Y direction, and MVPy indicates the component of the MVP in the Y direction. The following expressions related to the above-mentioned symbols all have the above-mentioned meanings.

For an extreme case, when TH=0, the 1/4 pixel accuracy range is reduced to zero, that is, the 1/2 pixel accuracy will be applied to all MVs except SKIP/DIRECT mode MVs. In this case, The MVP itself is rounded to half-pixel precision, as shown in equations (3)-(4):

MVPx=MVPx>>1<<1 (3)

MVPy＝MVPy>>1<<1 (4)

Due to the limitation of 1/4 pixel area, PMVR reduces the probability of 1/4 pixel MV being selected, thereby reducing the use of sub-pixel interpolation and reducing the complexity of the encoding and decoding process.

This step is taken for any motion search process, whether from unidirectional, symmetric or bidirectional encoding modes. In the motion search process, AVS2 adopts a step-by-step refinement method, first searches for 1/2 pixel precision points, and then searches for 1/4 pixel precision points. PMVR adjusts the search process of 1/4 pixel precision points. If the candidate 1/4 precision pixel point MV exceeds the range of 1/4 pixel precision, the MV is ignored and does not need to be searched, that is, the point is discarded. Do not participate in the decision of the best MV. This method eliminates part of the search for 1/4 pixel precision positions, which reduces the complexity of motion estimation.

Step S22, converting the MVs of the first area and the second area and representing according to the corresponding conversion.

In a specific implementation process, this step is used to calculate the MV Cost after the search is completed, or to encode the final MVD.

In this embodiment, for the MV in the first area, simply use the representation of MV-MVP to encode; Effective way to convert, not simply equal to MV–MVP. Because the MV components that exceed the range must be half-pixel precision, that is, they are even numbers in 1/4 pixel units, so the excess part can be doubled without losing the information of the MV.

Preferably, at the encoding end, for the MV of the second region, that is, the specific MV in this method, the MVD is adjusted using the Algorithm1 formula. The specific conversion process of the MVD is as follows:

Apply the conversion process described above to the motion vectors shown in Figures 3 and 4. Figure 3 shows the positions of A, B and C before conversion, and the positions of Figure 4 are obtained after converting the positions of A, B and C in Figure 3 through the above derivation process. In the actual operation process, the existing AVS2 standard is used to convert the motion vector shown in Figure 3, and the PMVR method described in this embodiment is used to convert the motion vector shown in Figure 3, and the MVD in the conversion result The values are listed in Table 1, as shown in Table 1:

Table 1

It can be seen from Table 1 that the value of MVD can be compressed into a smaller value through Algorithm 1 conversion of PMVR method, thereby reducing the number of bits for encoding MVD. It should be noted that when the motion vector MV is in the AVS2 encoding standard, the compressed MVD value is still encoded using the original CABAC entropy encoding model of AVS2, but the encoding amplitude becomes smaller.

Correspondingly, when decoding the compressed MV, that is, at the decoding end, for a specific MVD, that is, the MVD corresponding to the specific MV in the second area, the Algorithm 2 formula is used to restore or convert. When decoding each block, first read the MVD from the code stream, and then use the inversion algorithm corresponding to Algorithm 1 to obtain the MV instead of simple MVD+MVP. The specific conversion process is as follows:

The PMVR method of this embodiment is used for performance testing on the AVS2 reference software RD8.0. Table 2 lists the experimental results of PMVR on the AVS2 reference software RD8.0 (RA, TH=2). Among them, the test sequence used is the AVS2 standard test sequence, the first column is the video sequence resolution, and the second, third, and fourth columns are the BD-Rate performance results of the Y, U, and V components respectively. As shown in table 2:

Table 2

It can be seen from Table 2 that the compressed MVD value in this embodiment still adopts the original CABAC entropy coding model of AVS2 for coding, but the coding amplitude becomes smaller, which improves the compression effect of the MV amplitude, thereby reducing Encoding rate, so as to achieve the gain of encoding performance.

Fig. 5 is a schematic structural diagram of a motion vector compression device according to a third embodiment of the present invention.

As shown in FIG. 5, the motion vector compression device of this embodiment includes: a search unit 11 and a conversion unit 12, wherein,

The search unit 11 is used to search the pixel precision area of the motion vector, divide the motion vector into two areas, the two areas include a first area and a second area, and use different Search with pixel precision.

The converting unit 12 is connected to the searching unit 11, and is used to express the MV of the first area by using the first notation, and express the MV of the second area by using the second notation.

Further, the searching unit 11 is further configured to search the motion vector according to different areas of the first area with 1/4 pixel precision and the second area with 1/2 pixel precision.

Further, the search unit 11 is further configured to discard the motion vector corresponding to the motion vector when the searched motion vector is located in the second area of 1/2 pixel precision and the motion vector itself is 1/4 pixel precision. of pixels.

Further, the conversion unit 12 compresses the MV at the encoding end, and restores the MV at the decoding end.

Those skilled in the art can understand that the accompanying drawing is only a schematic diagram of an embodiment, and the modules or processes in the accompanying drawing are not necessarily necessary for implementing the present invention.

It can be seen from the above description of the implementation manners that those skilled in the art can clearly understand that the present invention can be implemented by means of software plus a necessary general hardware platform. Based on this understanding, the essence of the technical solution of the present invention or the part that contributes to the prior art can be embodied in the form of software products, and the computer software products can be stored in storage media, such as ROM/RAM, disk , CD, etc., including several instructions to make a computer device (which may be a personal computer, server, or network device, etc.) execute the methods described in various embodiments or some parts of the embodiments of the present invention.

The above is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed in the present invention can easily think of changes or Replacement should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be determined by the protection scope of the claims.

Claims (5)

1. a compression method of motion vector, it is characterized in that, described method comprises: Based on the AVS2 encoding standard, search the pixel precision area of the motion vector, use high pixel precision for MVs within a certain range close to the MVP, use low pixel precision for MVs within a certain range far away from the MVP, and convert the motion vector Divided into two areas of the first area of 1/4 pixel precision and the second area of 1/2 pixel accuracy, and only two areas, excluding the third area of 1/8 pixel accuracy, and to the The two areas use different pixel precision, adopt the method of step-by-step refinement, first search for 1/2 pixel precision points, then search for 1/4 pixel precision points, and realize the search through the threshold; expressing the motion vector of the first area by using the first notation, and expressing the motion vector of the second area by using the second notation; when the motion vector is located in the second area with 1/2 pixel precision and When the precision itself is 1/4 pixel, the pixel corresponding to the motion vector is discarded; by first searching for 1/2 pixel precision point, and then searching for 1/4 pixel precision point, when the candidate 1/4 precision pixel point MV If it exceeds the range of 1/4 pixel precision, the MV is ignored so that it does not need to be searched. This point is discarded and does not participate in the decision of the best MV. Part of the search for 1/4 pixel precision positions is eliminated, thereby reducing the cost of motion estimation. the complexity. 2. The motion vector compression method according to claim 1, wherein the motion vector of the second region is represented by a second representation, further, the motion vector is compressed at the encoding end, and the motion vector is compressed at the coding end. The decoding end restores the motion vector. 3. motion vector compression method according to claim 2, is characterized in that, described motion vector is compressed at encoding end, is further to adopt Algorithm1 formula to convert described motion vector, described at decoding end to The motion vector is restored, and the motion vector is further converted by using an Algorithm2 formula. 4. A compression device of a motion vector, characterized in that said device comprises: The search unit is used to search the pixel precision area of the motion vector, using high pixel precision for MVs within a certain range close to the MVP, using low pixel precision for MVs within a certain range far away from the MVP, and converting the motion vector Divided into two areas of the first area of 1/4 pixel accuracy and the second area of 1/2 pixel accuracy, and only two areas, excluding the third area of 1/8 pixel accuracy, and for both The area uses different pixel precision, and adopts the method of step-by-step refinement, first searches for 1/2 pixel precision points, then searches for 1/4 pixel precision points, and realizes the search through the threshold; and for discarding the pixel corresponding to the motion vector when the searched motion vector is located in the second area with 1/2 pixel precision and the motion vector itself is 1/4 pixel precision; And used, when the candidate 1/4 precision pixel point MV exceeds the range of 1/4 pixel precision, the MV is ignored so that it does not need to be searched, this point is discarded, and does not participate in the decision of the best MV, and a part of 1/4 4-pixel precision position search, thus reducing the complexity of motion estimation; A conversion unit, the conversion unit is connected to the search unit, and is configured to express the motion vector of the first area by using a first representation, and express the motion vector of the second area by using a second representation. 5. The device for compressing motion vectors according to claim 4, wherein the conversion unit uses a second representation to represent the motion vectors of the second region, and further comprises: at the encoding end, the motion vector The vector is compressed, and the motion vector is restored at the decoding end.