JP2015076766A - Video encoding device, video encoding method, and video encoding program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a video encoding device which, when a plurality of motion vector candidates exist, can broaden an actual search area for an encode block by adaptively correcting a search area and performing a search.SOLUTION: Provided is a video encoding device correcting a prediction vector candidate in order to avoid the duplication of a search range and generating a new prediction vector candidate before performing video encoding on the prediction vector candidate, the video encoding device comprising: center point detection means for calculating coordinates of a center point from among coordinates indicated by all motion vector candidates and outputting the calculated center point coordinates; detection means for detecting the overlapping of motion vector candidate search areas from the coordinates indicated by a plurality of motion vector candidates and a search range and outputting overlap information; and candidate vector correction means for correcting the value of the motion vector candidate using the coordinates indicated by the plurality of motion vector candidates, the coordinates of the center point, the overlap information, and reference image information, and outputting the corrected motion vector candidate.

Description

  The present invention relates to a video encoding device that performs video encoding using motion prediction, a video encoding method, and a video encoding program.

  MPEG-2, MPEG-4, and MPEG-4 / AVC are often used as video encoding techniques. In recent years, HEVC has been standardized as a video encoding standard following MPEG-4 / AVC. In the video coding standard, a method for reducing the amount of information using intra-screen coding, in which coding is performed using information closed in one frame, is common. In general, a method for reducing the amount of information is generally combined with inter-frame coding in which coding is performed using a plurality of temporally continuous frames.

  In inter-screen coding, motion prediction processing between different screens is performed to obtain motion vector information, and the amount of information is reduced by coding the difference values between the motion vector information and the screen. Here, it is possible to reduce the amount of information with high efficiency by capturing a correct motion and deriving a highly accurate motion vector. As a method for searching for a highly accurate motion vector, there is a full search method as the most basic method. The full search method searches for all areas in the screen from end to end, so that high accuracy can be realized, but the amount of calculation is enormous. Therefore, various methods have been proposed as a method for searching for a motion vector with high efficiency.

  The first is a method for searching around the prediction vector. This is a method based on the tendency that the motion prediction result of the encoded block and the motion prediction result of the surrounding encoded blocks have a high correlation, and MPEG-4 / AVC and HEVC have a method for calculating a prediction vector. For the motion vector of the block, the code amount is reduced by encoding only the difference value from the prediction vector. In this method, a prediction vector is calculated from the motion vectors already existing around the block using this prediction vector, and the periphery is searched using the position indicated by the prediction vector as the center of the search (for example, see Non-Patent Document 1).

  The second is a method for efficiently reducing search points. Rather than evaluating all search points in a certain search range as represented by step search, first, a wide range of discrete points are evaluated, and neighboring points are searched around the most suitable point. This is a method for determining a motion vector with a small amount of calculation. In recent years, a search method that combines step search and full search called EPZS (Enhanced Predictive Zonal Search) has been performed, and a search accuracy almost equal to that of full search is realized with a small amount of computation (for example, non-patented). Reference 2).

  Unlike a full search that searches all regions, the method of narrowing search points defines a plurality of candidate points in order to improve the accuracy of the search, and evaluates the candidates step by step. A configuration shown in FIG. 1 can be considered as one of such searchers. FIG. 1 is a diagram illustrating an example of a search range for motion vector candidates. In FIG. 1, mvA and mvB represent prediction vectors obtained from the motion prediction results of the upper and left neighboring encoded blocks, respectively, and the search range is ± 1 pixel from the center of the prediction vector. An n-pixel search is also conceivable. At this time, the reference image may have integer pixel accuracy, or may have decimal pixel accuracy using a reduced image. When the search shown in FIG. 1 is performed, a total of 9 pixels of 3 × 3 pixels are searched per search range. Therefore, if there are two candidates for the prediction vectors mvA and mvB, it is possible to search for 9 or more pixels. Become.

  From this, it is clear that it is beneficial to prepare a plurality of candidates as a technique for improving the accuracy of the search. Therefore, a method of preparing a plurality of motion vectors of surrounding encoded vectors and performing a search based on the motion vectors is considered (for example, see Non-Patent Document 2).

ISO / IEC 14496-2 (MPEG-4 recommendation) Alexis M. Tourapis. "Enhanced Predictive Zonal Search for Single and Multiple Frame Motion Estimation" .in Proceedings of Visual Communications and Image Processing, pp.1069-1079, January 2002.

  However, in the video encoding devices described in Non-Patent Documents 1 and 2, there is a possibility that overlap occurs in search areas for a plurality of motion vector candidates. When the search areas overlap, there is a problem that an actual search area per number of motion vector candidates is narrowed because the same search area is searched a plurality of times for each motion vector candidate.

  In the example shown in FIG. 1, the search areas of the prediction vectors mvA and mvB overlap. In such a case, there is a solution that the search processing is stopped in the overlapping search range, but when the time per predetermined processing such as hardware implementation is determined (here, given to the search processing) When the search process is stopped, an idle time during which no calculation is performed occurs. If the search process is stopped, there is a problem that an empty time during which the arithmetic unit does not perform processing occurs, and a sufficient search area cannot be searched.

  The present invention has been made in view of such circumstances. When there are a plurality of motion vector candidates, the search area is adaptively corrected to perform a search, whereby an actual search area for the encoded block is determined. An object is to provide a video encoding device, a video encoding method, and a video encoding program that can be widened.

  The present invention is a video encoding device that corrects a prediction vector candidate to avoid duplication of search ranges with respect to the prediction vector candidate, generates a prediction vector candidate newly, and performs video encoding. Center point detection means for calculating and outputting the coordinates of the center point from the coordinates indicated by the motion vector candidate, and detecting the overlap of the search area of the motion vector candidate from the coordinates and search range indicated by the plurality of motion vector candidates, Using the overlap detection means for outputting overlap information, the coordinates pointed to by a plurality of motion vector candidates, the coordinates of the center point, the overlap information, and reference image information, the value of the motion vector candidate is corrected, Candidate vector correction means for outputting a corrected motion vector candidate.

  The present invention is characterized in that the candidate vector correcting means corrects the value of the motion vector candidate by moving a coordinate indicated by the motion vector candidate in a horizontal direction.

  The present invention is characterized in that the candidate vector correcting means corrects the value of the motion vector candidate by moving a coordinate indicated by the motion vector candidate in a vertical direction.

  The present invention is characterized in that the candidate vector correcting means corrects the value of the motion vector candidate by changing the length of the vector of the motion vector candidate.

  The present invention is characterized in that the candidate vector correcting means corrects the value of the motion vector candidate by changing the direction of the vector of the motion vector candidate.

  The present invention is characterized in that the center point detecting means calculates the center point coordinates from the coordinate points indicated by all the motion vector candidates using an arithmetic mean.

  The present invention is characterized in that the center point detection means calculates a median value of coordinate points indicated by all the motion vector candidates as the center point coordinates.

  The present invention relates to a video encoding method performed by a video encoding apparatus that corrects a prediction vector candidate in order to avoid duplication of search ranges with respect to the prediction vector candidate, and newly generates a prediction vector candidate and performs video encoding. A center point detecting step for calculating and outputting the coordinates of the center point from the coordinates indicated by all the motion vector candidates, and the search area of the motion vector candidates from the coordinates and search ranges indicated by the plurality of motion vector candidates The motion vector candidate using the overlap detection step of detecting the overlap of the image and outputting the overlap information; the coordinates indicated by the plurality of motion vector candidates; the coordinates of the center point; the overlap information; and the reference image information. And a candidate vector correcting step for correcting the value of the motion vector and outputting a corrected motion vector candidate.

  The present invention is a video encoding program for causing a computer to function as the video encoding device.

  According to the present invention, information on a plurality of motion vector candidates is corrected using search range information, a new motion vector candidate is determined, and search range duplication does not occur in the newly determined motion vector candidate. As a result, it is possible to search for a wider range. Therefore, improvement in search performance can be expected, and encoding efficiency can be improved.

It is a figure which shows the search range example of a motion vector candidate. It is a block diagram which shows the structure of one Embodiment of this invention. It is a figure which shows the example of correction of a motion vector candidate. It is a figure which shows the example of correction of a motion vector candidate. It is a figure which shows the example of correction of a motion vector candidate. It is a figure which shows the example of correction of a motion vector candidate. It is a figure which shows the example of correction of a motion vector candidate. It is a figure which shows the example of correction of a motion vector candidate. It is a figure which shows the definition of the direction of a vector. It is a figure which shows the example of correction of a motion vector candidate. It is a figure which shows the example of correction of a motion vector candidate. It is a flowchart which shows the processing operation of the prediction vector production | generation part 100 shown in FIG.

  Hereinafter, a video encoding apparatus according to an embodiment of the present invention will be described with reference to the drawings. FIG. 2 is a block diagram showing the configuration of the embodiment. When the configuration of the video encoding device 1 is described with reference to FIG. 2, the well-known functions and configurations that the video encoding device normally has are explained and the configuration of the video encoding device 1 is not directly related to the description of the present invention. Illustration is omitted. The video encoding device 1 according to the present embodiment includes a prediction vector generation unit 100. The prediction vector generation unit 100 includes a center point detection unit 101, an overlap detection unit 102, and a candidate vector correction unit 103.

The center point detection unit 101 receives all motion vector candidates (motion vector candidates 1 to N) as input, and calculates a center point from the coordinate value indicated by each motion vector candidate. As a method for determining the center point, there are a method in consideration of an average value and a median value. For example, the method for calculating the center point as an average value is as follows. The number of candidate vectors is two, and mvA = (xA, yA) and mvB = (xB, yB), respectively. At that time, the center points (CenterX, CenterY) are as shown in equations (1) and (2).

Similarly, when it is determined by the median value, the equations (3) and (4) are obtained. Here, median (val1, val2,..., ValN) is a function that returns a central value when val1 to valN are arranged in ascending order. If N is an even number, the arithmetic average of the two middle values is returned.

The overlap detection unit 102 receives all the motion vector candidates (motion vector candidates 1 to N) and search range information as inputs, and detects the degree of overlap between the motion vectors. The degree of overlap is calculated from the search range (Sx, Sy) in addition to the coordinate value indicated by each motion vector candidate. For example, the number of candidate vectors is two and mvA = (xA, yA) and mvB = (xB, yB), respectively, and the overlap in the X direction at this time is OLx and the overlap in the Y direction is OLy. If there is an overlap here, equation (5) holds. At this time, OLx and OLy are determined as shown in equations (6) and (7), respectively.

On the other hand, if there is no overlap between candidate vectors, equation (8) is established. At this time, OLx and OLy are determined as shown in equations (9) and (10), respectively.

  Candidate vector correcting section 103 receives all motion vector candidates (motion vector candidates 1 to N), search range, center point, and reference image as input, corrects the motion vector candidate, and the search range of each motion vector overlaps. Make corrections so that no. There are several ways to make corrections. Here, the number of candidate vectors is 2, mvA = (xA, yA), mvB = (xB, yB), the overlap in the X direction is OLx, the overlap in the Y direction is OLy, and the search range is (Sx , Sy), and the case where the center point is (CenterX, CenterY) will be described as an example.

  The search range of each candidate vector before correction is shown in FIG. The shaded portion represents the search range by mvA and mvB, and the shaded portion represents the overlapping portion.

The first correction method is a horizontal slide. The horizontal slide is a method of sliding the coordinates indicated by the candidate vector in the horizontal direction. FIG. 3 and FIG. 4 show processing examples at the time of horizontal sliding. In FIG. 3, one candidate vector is fixed, and only the other candidate vector is slid in the horizontal direction. Various methods can be applied to select a vector to be slid. For example, a method of selecting the shorter or longer vector length, a method of selecting a vector close to the motion vector of the peripheral block, and the like can be applied. In FIG. 4, both candidate vectors are slid in the horizontal direction. The size to be slid is calculated from OLx representing the overlap in the X direction. The sliding direction is calculated from mvA, mvB, and the center point. When mvA is fixed and mvB is corrected, the sliding direction is determined by the sign of (xB-CenterX). When positive, slide in the plus direction; when negative, slide in the minus direction. If it is positive, correction is performed as shown in equation (11).

When correcting both mvA and mvB, the vector is corrected after a priority vector is determined in advance. When OLx is an even number, both xA and xB are corrected by adding only OLx / 2, but when OLx is an odd number, the size of the slide of the priority vector is increased. The sliding direction is calculated from mvA, mvB, and the center point. The moving directions of mvA and mvB are determined by the codes (xA-CenterX) and (xB-CenterX), respectively. When positive, slide in the plus direction; when negative, slide in the minus direction. When mvA is a priority vector and OLx is an odd number, when mvA slide direction is positive and mvB slide direction is negative and mvA and mvB are corrected, correction is made as shown in equations (12) and (13). Do.

The second correction method is a vertical slide. The vertical slide is a method of sliding the coordinates indicated by the candidate vector in the vertical direction. FIG. 5 and FIG. 6 show an example of processing during vertical sliding. In FIG. 5, one candidate vector is fixed, and only the other candidate vector is slid in the vertical direction. Various methods can be applied to select a vector to be slid. For example, a method of selecting the shorter or longer vector length, a method of selecting a vector close to the motion vector of the peripheral block, and the like can be applied. In FIG. 6, both candidate vectors are slid in the vertical direction. The size to be slid is calculated from OLy representing the overlap in the Y direction. The sliding direction is calculated from mvA, mvB, and the center point. When mvA is fixed and mvB is corrected, the sliding direction is obtained by the sign of (yB-CenterY). When positive, slide in the plus direction; when negative, slide in the minus direction. If it is positive, correction is performed as shown in equation (14).

When correcting both mvA and mvB, the vector is corrected after a priority vector is determined in advance. When OLy is an even number, both yA and yB are corrected by adding only OLy / 2. When OLy is an odd number, the size of the priority vector slide is increased. The sliding direction is calculated from mvA, mvB, and the center point. The moving directions of mvA and mvB are determined by the signs of (yA-CenterY) and (yB-CenterY), respectively. When positive, slide in the plus direction; when negative, slide in the minus direction. When mvA is a priority vector and OLy is an odd number, when mvA slide direction is positive and mvB slide direction is negative and mvA and mvB are corrected, correction is made as shown in equations (15) and (16). Do.

  The third correction method is vector expansion / contraction. Vector expansion / contraction is a method of changing the length of the coordinate pointed to by the candidate vector with the same vector direction. 7 and 8 show examples of processing at the time of vector expansion / contraction. In FIG. 7, one candidate vector is fixed and only the other candidate vector is expanded or contracted. Various methods can be applied to select a vector to be expanded or contracted. For example, a method of selecting the shorter or longer vector length, a method of selecting a vector close to the motion vector of the peripheral block, and the like can be applied. In FIG. 8, both candidate vectors are expanded and contracted. FIG. 9 is a diagram showing the definition of the vector direction. The size of expansion / contraction is obtained from mvA, mvB, OLx, and OLy. The size of expansion / contraction is calculated from OLx representing the overlap in the X direction and OLy representing the overlap in the Y direction. Specifically, when the direction of the vector to be expanded / contracted is upward or downward, calculation is made from OLx, and when the direction of the vector to be expanded / contracted is rightward or downward, it is calculated from OLy. The direction of expansion / contraction is calculated from mvA, mvB, and the center point.

When mvA is fixed and mvB is corrected, the direction of expansion / contraction is compared with the length of the candidate vector and the length of the vector from the origin to the center point (hereinafter referred to as the center point vector). If the candidate vector is longer than the center point vector, the vector is corrected in the direction of extending the candidate vector, and if the candidate vector is shorter than the center point vector, the vector is corrected in the direction of reducing the candidate vector. When it is in the extending direction, correction is performed as shown in equation (17). However, at this time, | xB | is negative, | yB | is positive, and mvB is upward. Here, the result of xA × OLy / yA is rounded up.

When correcting both mvA and mvB, the vector is corrected after a priority vector is determined in advance. When the value of OLx or OLy corresponding to the direction of expansion / contraction is an even number, the correction is performed by adding 1/2 of the value, but when the value of OLx or OLy is an odd number, the size of expansion / contraction of the priority vector is set. Enlarge. The direction of expansion / contraction is compared with the length of the candidate vector and the length of the center point vector. If the candidate vector is longer than the center point vector, the vector is corrected in the direction of extending the candidate vector, and if the candidate vector is shorter than the center point vector, the vector is corrected in the direction of reducing the candidate vector. When mvA is a priority vector, mvA is corrected in the downward direction, and mvB is corrected in the upward direction, mvA and mvB are corrected as shown in equations (18) and (19). However, at this time, | xA | is negative, | yA | is positive, | xB | is negative, and | yB | is positive. Here, the results of xA × OLy / 2yA, yB × OLy / 2yB, and OLy / 2 are rounded up.

The fourth correction method is vector rotation. Vector rotation is a method of changing the direction of vectors so as to increase the distance between candidate vectors. 10 and 11 show processing examples during vector rotation. In FIG. 10, one candidate vector is fixed and only the other candidate vector is rotated. Various methods can be applied as a method for selecting a vector to be rotated. For example, a method of selecting the shorter or longer vector length, a method of selecting a vector close to the motion vector of the peripheral block, and the like can be applied. In FIG. 11, both candidate vectors are rotated. The direction of the vector for rotation is obtained from mvA, mvB, and the center point. When mvB is moved, it is obtained from the vector direction from mvA or the center point to mvB. The magnitude of the vector movement is obtained from the vector direction and OLx and OLy.
When mvA is fixed and mvB is corrected, correction is performed as shown in equation (20). At this time, it is assumed that the direction of rotation is the right direction. The result of yA × OLy / xA is rounded up.

When correcting both mvA and mvB, the vector is corrected after a priority vector is determined in advance. When the value of OLx or OLy corresponding to the direction of rotation is an even number, the correction is performed by adding 1/2 of that value, but when the value of OLx or OLy is an odd number, the magnitude of movement of the priority vector is set. Enlarge. The direction of the vector for rotation is obtained from mvA, mvB, and the center point. When mvA is a priority vector, mvA is rotated to the left, and mvB is rotated to the right, correction is performed as shown in equations (21) and (22). At this time, the results of yA × OLy / 2xA, yB × OLy / 2xB, and OLy / 2 are rounded up.

  Next, the processing operation of the prediction vector generation unit 100 shown in FIG. 2 will be described with reference to FIG. FIG. 12 is a flowchart showing the processing operation of the prediction vector generation unit 100 shown in FIG. First, the center point detection unit 101 receives all motion vector candidates (motion vector candidates 1 to N), and detects a center point from all the motion vector candidates (step S201).

  Next, the overlap detection unit 102 inputs all motion vector candidates (motion vector candidates 1 to N) and the search range, and detects an overlap from the motion vector candidates (step S202).

  Next, the overlap detection unit 102 determines the presence or absence of overlap from the output result of overlap detection (step S203). As a result of this determination, if there is an overlap, the candidate vector correction unit 103 corrects the candidate vector and corrects the vector information so that there is no overlap between the candidate vectors (step S204).

  On the other hand, if there is no overlap as a result of the determination, or after the candidate vectors have been corrected, the overlap detection unit 102 determines whether all the motion vector candidates have been processed so that there is no overlap (step) S205). As a result of the determination, if all the candidates have not been processed, the process returns to step S202 to repeat the process.

  On the other hand, when the process has been completed for all candidate vectors, the candidate vector correcting unit 103 outputs all the corrected candidate vectors and completes the process (step S206).

  As described above, when there are a plurality of motion prediction vectors for determining the search range, the search range for the coding block is set wide by correcting the motion prediction vector so that the search range does not overlap. It is possible to provide an algorithm suitable for hardware implementation.

  The prediction vector generation unit 100 in the embodiment described above may be realized by a computer. In that case, a program for realizing this function may be recorded on a computer-readable recording medium, and the program recorded on this recording medium may be read into a computer system and executed. Here, the “computer system” includes an OS and hardware such as peripheral devices. The “computer-readable recording medium” refers to a storage device such as a flexible medium, a magneto-optical disk, a portable medium such as a ROM and a CD-ROM, and a hard disk incorporated in a computer system. Furthermore, the “computer-readable recording medium” dynamically holds a program for a short time like a communication line when transmitting a program via a network such as the Internet or a communication line such as a telephone line. In this case, a volatile memory inside a computer system serving as a server or a client in that case may be included and a program held for a certain period of time. Further, the program may be for realizing a part of the functions described above, and may be a program capable of realizing the functions described above in combination with a program already recorded in the computer system. It may be realized using hardware such as PLD (Programmable Logic Device) or FPGA (Field Programmable Gate Array).

  As mentioned above, although embodiment of this invention has been described with reference to drawings, the said embodiment is only the illustration of this invention, and it is clear that this invention is not limited to the said embodiment. is there. Therefore, additions, omissions, substitutions, and other modifications of the components may be made without departing from the technical idea and scope of the present invention.

  When there are a plurality of motion vector candidates, the search area is adaptively corrected and the search is performed, so that it can be applied to an application in which it is essential to widen the actual search area for the coding block.

  DESCRIPTION OF SYMBOLS 1 ... Video coding apparatus, 100 ... Predictive vector production | generation part, 101 ... Center point detection part, 102 ... Overlap detection part, 103 ... Candidate vector correction part

Claims (9)

A prediction vector candidate is a video encoding device that corrects a prediction vector candidate in order to avoid duplication of search ranges, newly generates a prediction vector candidate, and performs video encoding,
Center point detection means for calculating and outputting the coordinates of the center point from the coordinates indicated by all motion vector candidates;
An overlap detection means for detecting an overlap of search areas of the motion vector candidates from coordinates and search ranges indicated by the plurality of motion vector candidates, and outputting overlap information;
A candidate that corrects the value of the motion vector candidate using the coordinates indicated by the plurality of motion vector candidates, the coordinates of the center point, the overlap information, and reference image information, and outputs a corrected motion vector candidate A video encoding device comprising: vector correction means.   The video encoding apparatus according to claim 1, wherein the candidate vector correcting unit corrects the value of the motion vector candidate by moving a coordinate indicated by the motion vector candidate in a horizontal direction.   The video encoding apparatus according to claim 1, wherein the candidate vector correcting unit corrects the value of the motion vector candidate by moving a coordinate indicated by the motion vector candidate in a vertical direction.   The video encoding apparatus according to claim 1, wherein the candidate vector correcting unit corrects the value of the motion vector candidate by changing a length of the vector of the motion vector candidate.   The video encoding apparatus according to claim 1, wherein the candidate vector correcting unit corrects the value of the motion vector candidate by changing a direction of the vector of the motion vector candidate.   6. The video according to claim 1, wherein the center point detecting means calculates the center point coordinates from the coordinate points indicated by all the motion vector candidates using an arithmetic mean. Encoding device.   6. The video encoding device according to claim 1, wherein the center point detection unit calculates, as the center point coordinates, median values of coordinate points indicated by all the motion vector candidates. 7. . A predictive vector candidate is a video encoding method performed by a video encoding device that corrects a predictive vector candidate in order to avoid duplication of search ranges, newly generates a predictive vector candidate, and performs video encoding,
A center point detection step of calculating and outputting the coordinates of the center point from the coordinates indicated by all motion vector candidates;
An overlap detection step of detecting an overlap of search areas of the motion vector candidates from coordinates and search ranges indicated by the plurality of motion vector candidates, and outputting overlap information;
A candidate that corrects the value of the motion vector candidate using the coordinates indicated by the plurality of motion vector candidates, the coordinates of the center point, the overlap information, and reference image information, and outputs a corrected motion vector candidate A video encoding method comprising: a vector correction step.   A video encoding program for causing a computer to function as the video encoding device according to any one of claims 1 to 7.

Images