CN102917225B - HEVC intraframe coding unit fast selecting method - Google Patents

Abstract

The invention provides a kind of HEVC intraframe coding unit fast selecting method.Relevant with the histogram of gradients of the non-normalized of CU owing to being similar to absolute error and SAD the rate distortion costs of each CU obtained.The CU of two equal dimensions, its non-normalized histogram of gradients similarity degree is higher, and segmentation identifier is more equal.The present invention uses the template matches mode based on non-normalized histogram of gradients to realize the quick selection of CU, thus improves the method for the coding rate of HEVC intra-frame prediction method.Utilize non-normalized histogram of gradients as feature to judge that each CU is the need of being divided into four less CU.Under the condition meeting template matches, eliminate the step comparing encoding rate distortion cost one by one.

Description

HEVC intraframe coding unit fast selecting method

Technical field

The present invention propose in high-performance video coding (High Efficiency Video Coding, HEVC) technology, particularly in coding to the technology that predict coding unit is selected.

Background technology

Along with development and the maturation of high-definition video technology, in people's daily life, to emerge in large numbers the application of a large amount of HD videos.Various application about HD video receive welcome and the favor of user.But, higher resolution bring larger data volume to make the compression performance of HD video to coding and decoding video algorithm require higher.Current video encoding and decoding standard (as H.264/AVC etc.) often can not meet the requirement of HD video to compression performance well.For meeting the requirement of new Video Applications to encoding and decoding standard, the Video coding combination and cooperation group (Joint Collaborative Team on Video Coding, JCT-VC) of ITU-T/ISO/IEC is formulating new video encoding and decoding standard HEVC of future generation a few days ago.For being applicable to the real-time application of video technique, we need the coding rate improving HEVC.The present invention is mainly devoted to the intraframe coding speed improving HEVC, reduces computation complexity.

At present, the fast algorithm of a class HEVC intraframe coding is mainly contained.Namely based on the quick selection of prediction direction.This method has and judges prediction direction by the gradient information of current block; Also have and utilize the prediction direction of adjacent block to carry out the prediction direction of auxiliary judgment to current block.But these methods are that the quick selection by realizing prediction direction improves intraframe coding speed mostly.Do not attempt going to deal with problems from the angle of the quick selection of coding unit (CU, Coding Unit).In fact, during intraframe coding, the yardstick of CU has 64x64,32x32,16x16,8x8.Time the maximum coding unit (LCU) of each 64x64 is encoded, need down to travel through according to quad-tree structure from yardstick 64x64 to 8x8 to carry out rate-distortion optimization, choose most suitable CU yardstick.Smallest dimension due to CU is 8x8, can not down be divided into less CU again for 8x8CU.But the predicting unit (Prediction Unit, PU) that can be divided into 4 4x4 of 8x8 CU is encoded.If we fast prediction can go out final intraframe coding CU yardstick, in intra-prediction process, so just can save the forecasting process of a lot of CU.

In existing HEVC intraframe coding process, be by traveling through from 64x64 (degree of depth is minimum) to the CU of 8x8 yardstick according to quad-tree structure to the coding of each LCU.Each CU has one to split identifier (splitflag) in ergodic process.This splitflag is used for judging that current C U is the need of the CU being divided into four more small scales, namely calculates and the rate distortion costs J after segmentation split coding of CU under comparing current size _splitwith the rate distortion costs J do not split unsplit and encode _unsplitsize; As rate distortion costs J _unsplitbe less than or equal to rate distortion costs J _split, then current CU is no longer split, and the segmentation identifier marking this CU is unsplit; As rate distortion costs J _unsplitbe greater than rate distortion costs J _splitthen current CU also needs to split, the segmentation identifier marking this this CU is split, and continue utilization rate distortion cost judge this CU split after whether also need further segmentation, until judge complete to smallest dimension 8x8 (degree of depth is maximum), while the segmentation identifier of current C U is determined, the coded message of preserving mode corresponding to segmentation identifier upgrades the corresponding coded message of current LCU.

Summary of the invention

Technical problem to be solved by this invention is, a kind of method of the quick selection HEVC intraframe coding unit based on template matches.

The technical scheme that the present invention adopts for the above-mentioned technology of solution is, HEVC intraframe coding unit fast selecting method, comprises the following steps:

In step 1, the maximum coding unit LCU to be predicted that extracts successively in current encoded frame, from the degree of depth corresponding to maximum coding yardstick, the judgement of coding unit yardstick is carried out this to LCU;

Step 2, kth the coding unit of calculating current encoded frame under current depth i non-normalized histogram of gradients, judge segmentation identifier encoded under current depth i in current encoded frame be segmentation split coding unit number and segmentation identifier is the coding unit number not splitting unsplit whether be all more than or equal to default template matches and start thresholding N, as no, the coding unit yardstick carried out based on rate distortion costs judges, namely enters step 3; In this way, the coding unit yardstick carried out based on template matches judges, namely enters step 4;

Step 3, coding unit yardstick determining step based on rate distortion costs: to the coding unit under current depth i carry out unsplit mode coding respectively, split mode encodes, compare the rate distortion costs size of two kinds of coded systems, the mode that selection rate distortion cost is less is segmentation identifier assignment; Work as coding unit segmentation identifier assignment be unsplit, then enter step 5; As coding unit segmentation identifier assignment be split, then enter step 6;

Step 4, coding unit yardstick determining step based on template matches: by coding unit non-normalized histogram of gradients mate with the non-normalized histogram of gradients of coding unit encoded under current depth i in current encoded frame, find out and coding unit the individual encoded coding unit of the most similar setting number x; Judge that whether the segmentation identifier of the individual encoded coding unit of x is all identical, as no, return step 3; Coding unit as encoded in x segmentation identifier be unsplit, then to coding unit carry out unsplit mode to encode and calculation rate distortion cost under corresponding coded system, then enter step 5; Coding unit as encoded in x segmentation identifier be split, to coding unit carry out split mode coding and rate distortion costs under calculating corresponding coded system, then enter step 6; Wherein, x≤N;

Step 5, retain the rate distortion costs and coded message of encoding in unsplit mode, upgrading in current encoded frame encoded segmentation identifier under current depth i is the coding unit number of unsplit initial value is 0;

Step 6, retain the rate distortion costs and coded message of encoding in split mode, upgrading in current encoded frame encoded segmentation identifier under current depth i is the coding unit number of split initial value is 0; Judge that whether current depth i is the degree of depth that coding yardstick 8x8 is corresponding, if current depth is the degree of depth that coding yardstick 8x8 is corresponding, then judge the coding unit whether current LCU needs yardstick in addition and judge under current depth further, in this way, upgrade k=k+1, return step 2, as no, enter step 7; If current depth is not the degree of depth that 8x8 is corresponding, then upgrade current depth i=i+1, return step 2;

Step 7, to judge in current LCU whether also have the uncoded coding unit completed, in this way, renewal current depth i is the degree of depth that the yardstick of the next uncoded coding unit completed is corresponding, as no, current LCU has encoded, and judges whether the postamble arriving current encoded frame further, as arrived postamble, then current encoded frame end-of-encode, as do not arrived postamble, returns step 1.

The present invention proposes the quick selection that a kind of template matches mode based on non-normalized histogram of gradients realizes CU, thus improves the method for the coding rate of HEVC intra-frame prediction method.

Find by analysis, the rate distortion costs being similar to each CU obtained with absolute error and SAD is directly proportional to the difference of predict pixel and source pixel, and this proportional relation is relevant with the non-normalized histogram of gradients of CU.The CU of two equal dimensions, its non-normalized histogram of gradients similarity degree is higher, and segmentation identifier is more equal.Therefore, the present invention utilizes non-normalized histogram of gradients as feature to judge that each CU is the need of being divided into four less CU.Under the condition meeting template matches, eliminate the step comparing encoding rate distortion cost one by one.

The invention has the beneficial effects as follows, the mode of template matches is adopted to predict the partitioned mode of current C U based on non-normalized histogram, computation complexity is low, Detection accuracy is high, method is simple and efficiently, ensureing the time complexity significantly reducing HEVC infra-frame prediction under the prerequisite that coding quality loses hardly.

Accompanying drawing explanation

Fig. 1 is template matches fast prediction mode example.

Embodiment

The present invention utilizes non-normalized histogram of gradients as feature to judge that current C U is the need of being divided into four less CU.Consider that, in same frame of video, similar region may be repeated.In the CU of the same scale that we adopt a kind of mode of template matches encoded before present frame, find the non-normalized the most similar x of the histogram of gradients CU to current C U, judged the splitflag of current C U by the corresponding splitflag of this x CU.When the splitflag homogeneous phase of this x CU while, the splitflag of current C U is also equal with the splitflag of this x CU.Otherwise when this x CU is incomplete same, current C U judges at partitioning scheme as split or unsplit according to existing rate distortion costs manner of comparison.

As shown in Figure 1, black block cu _kfor the current CU encoded.C={cu ₀, cu ₁..., cu _k-1be the set of the encoded CU of present frame, be made up of grey block and white blocks.Wherein grey block is split block, and white blocks is unsplit block.The process of template matches is exactly search and cu in C _kx the most similar CU.If this x CU for splitflag identical, so cu _kalso identical splitflag is adopted.Otherwise, calculate the rate distortion costs of current C U in split and unsplit situation, select splitflag by comparing rate distortion cost.Namely

${\mathrm{splitflag}}_{{\mathrm{cu}}_k}=\left\{\begin{array}{cc}\mathrm{unsplit}& x_{\mathrm{mu}}==x\left|\right|J_{\mathrm{unsplit}}\&le;J_{\mathrm{split}}\\ \mathrm{split}& x_{\mathrm{ms}}==x\left|\right|J_{\mathrm{split}}<J_{\mathrm{unsplit}}\end{array}\right.---\left(1\right)$

Wherein x _muand x _msbe respectively the CU number of unsplit and split in x the CU mated most.J _splitand J _unsplitbe respectively the rate distortion costs that current block calculates according to split and unsplit mode.

In addition, concerning the CU of each degree of depth i, if the encoded CU of present frame only belongs to a class, then can not distinguish current block well.Because the similar CU found when coupling also belongs to this class.Go down successively, the CU of whole frame belongs to this class.Obviously predicated error is larger like this.So required before template matches, two classes must be not less than a definite value with reference to minimum value in the number of CU, and the template matches namely preset starts thresholding N:

$\min (x_{\mathrm{pu}}^i,x_{\mathrm{ps}}^i)\&GreaterEqual;N---\left(2\right)$

Wherein, with be respectively unsplit CU encoded in present frame and the number of split CU.

Due to more for the CU number compared with small scale in high-definition image, for making matching effect better, a kind of mode of similar uniform sampling can be adopted to carry out palette sample coupling according to certain step-length s to CU, namely every s encoded CU samples one and compares.In order to make the value of step-length s more flexible, step-length s is dynamic step length, and the size of s is relevant with CU number encoded in present frame:

$s=\max (1,\frac{x_{\mathrm{pu}}^i+x_{\mathrm{ps}}^i}{T})---\left(3\right)$

T is a controling parameters, can rule of thumb regulate.

Embodiment

The present invention realizes on HM6.0 experiment porch, and step is as follows:

Initialization step:

1st step: in the maximum coding unit LCU to be predicted extracted successively in current encoded frame, carries out coding unit yardstick judgement to LCU this from the degree of depth corresponding to maximum coding yardstick; Before the CU of each degree of depth i of each frame coded frame is encoded, with be initialized as 0;

Coding unit yardstick determining step:

2nd step: calculate kth the coding unit of current encoded frame under current depth i non-normalized histogram of gradients; Judge that segmentation identifier encoded under current depth i in current encoded frame is the coding unit number of segmentation split and segmentation identifier is the coding unit number not splitting unsplit whether be all more than or equal to default template matches and start thresholding N, namely judge whether to meet $\min (x_{\mathrm{pu}}^i,x_{\mathrm{ps}}^i)\&GreaterEqual;N$ And $\max (x_{\mathrm{pu}}^i,x_{\mathrm{ps}}^i)\&GreaterEqual;x,$ As no, the coding unit yardstick carried out based on rate distortion costs judges, namely enters step 3; In this way, the coding unit yardstick carried out based on template matches judges, namely enters step 4;

Step 3, coding unit yardstick determining step based on rate distortion costs: to the coding unit under current depth i carry out unsplit mode coding respectively, split mode encodes, compare the rate distortion costs size of two kinds of coded systems, the mode that selection rate distortion cost is less is segmentation identifier assignment; Work as coding unit segmentation identifier assignment be unsplit, then enter step 5; As coding unit segmentation identifier assignment be split, then enter step 6;

Step 4, coding unit yardstick determining step based on template matches: by coding unit non-normalized histogram of gradients with sample current encoded frame every step-length s in the non-normalized histogram of gradients of coding unit encoded under current depth i mate, find out and coding unit the individual encoded coding unit of the most similar setting number x; Judge the coding unit that x is encoded segmentation identifier whether all identical, i.e. whether max (x _mu, x _ms)==x, as no, returns step 3; Coding unit as encoded in x segmentation identifier be unsplit, then to coding unit carry out unsplit mode to encode and calculation rate distortion cost under corresponding coded system, then enter step 5; Coding unit as encoded in x segmentation identifier be split, to coding unit carry out unsplit mode coding and rate distortion costs under calculating corresponding coded system, then enter step 6; Wherein, x≤N;

Cycle criterion step:

Step 5, retain the rate distortion costs and coded message of encoding in unsplit mode, upgrading in current encoded frame encoded segmentation identifier under current depth i is the coding unit number of unsplit

Step 6, retain the rate distortion costs and coded message of encoding in split mode, upgrading in current encoded frame encoded segmentation identifier under current depth i is the coding unit number of split judge that whether current depth i is the degree of depth that coding yardstick 8x8 is corresponding, if current depth is the degree of depth that coding yardstick 8x8 is corresponding, then judge the coding unit whether current LCU needs yardstick in addition and judge under current depth further, in this way, upgrade k=k+1, return step 2, as no, enter step 7; If current depth is not the degree of depth that 8x8 is corresponding, then upgrade current depth i=i+1, return step 2;

Step 7, to judge in current LCU whether also have the uncoded coding unit completed, in this way, renewal current depth i is the degree of depth that the yardstick of the next uncoded coding unit completed is corresponding, as no, current LCU has encoded, and judges whether the postamble arriving current encoded frame further, as arrived postamble, then current encoded frame end-of-encode, as do not arrived postamble, returns step 1.

Claims (3)

1.HEVC intraframe coding unit fast selecting method, is characterized in that, comprise the following steps:

In step 1, the maximum coding unit LCU to be predicted that extracts successively in current encoded frame, from the degree of depth corresponding to maximum coding yardstick to carrying out the judgement of coding unit yardstick this LCU;

A kth coding unit in step 2, calculating current encoded frame under current depth i non-normalized histogram of gradients, judge segmentation identifier encoded under current depth i in current encoded frame be segmentation split coding unit number and segmentation identifier is the coding unit number not splitting unsplit whether be all more than or equal to default template matches and start thresholding N, as no, the coding unit yardstick carried out based on rate distortion costs judges, namely enters step 3; In this way, the coding unit yardstick carried out based on template matches judges, namely enters step 4;

Step 3, coding unit yardstick determining step based on rate distortion costs: to the coding unit under current depth i carry out unsplit mode coding respectively, split mode encodes, compare the rate distortion costs size of two kinds of coded systems, the mode that selection rate distortion cost is less is segmentation identifier assignment; Work as coding unit segmentation identifier assignment be unsplit, then enter step 5; As coding unit segmentation identifier assignment be split, then enter step 6;

Step 4, coding unit yardstick determining step based on template matches: by coding unit non-normalized histogram of gradients mate with the non-normalized histogram of gradients of coding unit encoded under current depth i in current encoded frame, find out and coding unit the individual encoded coding unit of the most similar setting number x; Judge that whether the segmentation identifier of the individual encoded coding unit of x is all identical, as no, return step 3; Coding unit as encoded in x segmentation identifier be unsplit, then to coding unit carry out unsplit mode to encode and calculation rate distortion cost under corresponding coded system, then enter step 5; Coding unit as encoded in x segmentation identifier be split, to coding unit carry out split mode coding and rate distortion costs under calculating corresponding coded system, then enter step 6; Wherein, x≤N;

Step 5, retain the rate distortion costs and coded message of encoding in unsplit mode, upgrading in current encoded frame encoded segmentation identifier under current depth i is the coding unit number of unsplit initial value is 0;

Step 6, retain the rate distortion costs and coded message of encoding in split mode, upgrading in current encoded frame encoded segmentation identifier under current depth i is the coding unit number of split initial value is 0; Judge that whether current depth i is the degree of depth that coding yardstick 8x8 is corresponding, if current depth is the degree of depth that coding yardstick 8x8 is corresponding, then judge the coding unit whether current LCU needs yardstick in addition and judge under current depth further, in this way, upgrade k=k+1, return step 2, as no, enter step 7; If current depth is not the degree of depth that 8x8 is corresponding, then upgrade current depth i=i+1, return step 2;

Step 7, to judge in current LCU whether also have the uncoded coding unit completed, in this way, renewal current depth i is the degree of depth that the yardstick of the next uncoded coding unit completed is corresponding, as no, current LCU has encoded, and judges whether the postamble arriving current encoded frame further, as arrived postamble, then current encoded frame end-of-encode, as do not arrived postamble, returns step 1.

2. HEVC intraframe coding unit fast selecting method as claimed in claim 1, is characterized in that, in described step 4, by coding unit non-normalized histogram of gradients and current encoded frame under current depth i, mate by the sample non-normalized histogram of gradients of encoded coding unit of dynamic step length, described dynamic step length value is directly proportional to encoded coding unit number.

3. HEVC intraframe coding unit fast selecting method as claimed in claim 2, it is characterized in that, the concrete grammar of described dynamic step length value is: wherein s is dynamic step length value, and T is controling parameters.