CN103905818A - Method for rapidly determining inter-frame prediction mode in HEVC standard based on Hough conversion - Google Patents

Abstract

The invention discloses a method for rapidly determining an inter-frame prediction coding mode in the latest national video compression standard, namely, the HEVC standard. The method includes the steps of firstly, conducting edge detection and Hough conversion on a current PU; secondly, conducting statistic analysis on the tangent value of the direction angle of a detected straight line so as to extract the most possible edge direction of the current PU according to the heuristic type information, optimally selecting candidate prediction modes so as to determine the MPM by conducting detection through the code rate and RMD process, and determining the final coding mode by conducting detection through the code rate and RDO process. Experiments show that the running speed of the HEVC can be increased by more than 20% according to the method.

Description

Intra prediction mode fast determination method in HEVC standard based on Hough transformation

Technical field

The present invention relates in field of video encoding at present the method for accelerating its speed of service in up-to-date in the world compression standard HEVC, relate in particular to a kind of definite method that adopts the technology such as Hough (Hough) conversion to accelerate the optimum code pattern of its Intra-coded blocks.

Background technology

HEVC (High Efficiency Video Coding) is current up-to-date international video encoding standard.It moves motion picture expert group version (MPEG) by ISO/IEC and ITU-T Video coding expert group (VCEG) combines formulation.It is the high-performance video coding standard after H.264/AVC.Its code efficiency is than H.264 having improved more than 50%.HEVC still adopts prediction to add the hybrid encoding frame of conversion, technology in this framework technology before relatively, although coding framework content and flow process roughly do not have too much change, in order to improve Video coding efficiency, many new technology in HEVC standard, are also added.

In video encoding standard H.264, macro block (Micro Block, MB) is the most basic unit of coding.But in HEVC coding standard, due to popularizing of HD video, to reaching the super clear video of 2K, 4K, if adopt again the encoding block of such fixed size, be difficult to adapt to the demand of HD video.Therefore, HEVC does not continue to adopt the concept of MB, represents: coding unit (CU coding unit), predicting unit (PU prediction unit), converter unit (TU transform unit) but adopt more flexi mode to carry out block structure.CU is the most basic cutting unit, and its maximum coding size is 64 × 64, and minimum code size is 8 × 8.PU is one and carries and the elementary cell of forecasting process relevant information, and PU size is less than or equal to CU size.A CU can comprise the predicting unit PU of one or more different sizes, and a PU comprises some converter unit TU, and TU is an elementary cell based on conversion and quantification.HEVC adopts quad-tree structure to carry out the division of coding unit.

HEVC is for shortcomings such as predictive mode number in having overcome are H.264/AVC few, precision of prediction is not high, inaccurate, and HEVC has increased the quantity of adoptable predictive mode greatly.HEVC infra-frame prediction has 35 kinds of predictive modes, and wherein pattern 0 represents to use plane (Planar) mode to predict, pattern 1 represents to use direct current (DC) mode to predict, pattern 2 to 34 expressions are used various angles to predict.

In the reference model HM4.0 of HEVC, the decision process of frame mode is divided into three phases.First, the set of patterns that is applicable to different size PU is roughly selected, calculate the rate distortion costs of predictive mode (Rate-Distortion Cost likely, Rd-Cost), and it is arranged according to cost incremental order, one group of minimum predictive mode of Rd-Cost value is to roughly select subset, and this process is called mode decision (RMD) process of roughly selecting.Here, when calculated distortion, only consider the distortion of prediction residual, instead of distortion after encoding and decoding.Secondly, whether the intra prediction mode of checking current block left side and upside reference block is included in is roughly selected in subset, if do not exist, is added subset.Finally, the predictive mode of roughly selecting in subset is carried out to rate-distortion optimization, the pattern of Least-cost is optimal prediction modes, and this process is called code check and distortion factor optimization (RDO) process.Here, need to calculate the distortion after the encoding and decoding of each candidate pattern in RDO process, therefore its amount of calculation is very large.Owing to having the pattern of a lot of optional infra-frame predictions in HEVC, the deterministic process of its frame mode has very large computation complexity, very consuming time.

In sum, along with the quantitative increase of optional predictive mode in frame in HEVC, although improved precision of prediction and code efficiency, also greatly increased the encoder complexity of HEVC simultaneously.Thereby optimal prediction modes is really fixed time and has been increased widely in frame.

The present invention makes every effort to keeping on the basis of precision of prediction, reduces intra prediction mode in HEVC and really fixes time.Because the marginal information of the straight line in image block can be brought the direction of the tendency of pixel in piece to us, the present invention will utilize this heuristic information to simplify and reduce further candidate's intra prediction mode.

Summary of the invention

In view of the above-mentioned defect of prior art, the present invention proposes a kind of definite new method of the fast intra mode based on Hough conversion.For realizing the object that keeps as much as possible the precision of prediction in reducing amount of calculation, first the present invention will carry out rim detection to image, then adopt Hough conversion to search the edge for straight line, next determine the angle of the linear edge in present image predicting unit PU, utilize statistic histogram to determine the prediction direction that current PU is most possible, and according to the most possible candidate's of this orientation determination predictive mode, with roughly selecting the candidate pattern that in mode decision (RMD) process RMD and code check and distortion factor optimization (RDO) process, minimizing will detect, reduce widely amount of calculation, and the predictive mode simultaneously current PU being chosen as much as possible.It is characterized in that, for the each intraprediction unit PU in video image, described method comprises:

Step 1, to current PU, judges whether its size is 4 × 4.In this way, carry out step-length and be 1 pattern sub-sampling, and leap to step 8.Otherwise, utilize Canny operator to carry out computing to current PU.

Step 2, according to the result of the computing of Canny operator, carries out refinement with morphologic method edge.

Step 3, selects suitable threshold value to carry out extraction and the connection of marginal point in PU.

Step 4, to current PU, utilizes Hough change detection linear edge pixel,, retains the marginal point on a certain straight line that is, removes the marginal point on curve.

Step 5, to each linear edge, determines the differential seat angle θ in itself and horizontal direction, and its tangent value tan θ.

Step 6, according to the scope of the tangent value of the pre-measuring angle of each intra prediction mode, with the tangent value of each linear edge in the current PU detecting, determines and in current PU, drops on the linear edge pixel sum within the scope of each with statistic histogram.

Step 7, according to this statistic histogram, determines several predictive modes of the linear edge pixel sum maximum in histogram.

Step 8, according to the size of current PU, and the statistic histogram obtaining, definite candidate's who enters pattern rougher process RMD frame mode.

Step 9, according to RMD process, and the size of current PU, definite candidate's who enters pattern refining process RDO frame mode.

Step 10, according to the result of RDO, determines the intra prediction mode of the optimization of current PU.

Further, in described step 1, the piece to 4 × 4 carries out pattern sub-sampling.In HEVC reference model HM4.0, for totally 19 kinds of the applicable intra prediction modes of 4 × 4PU, directivity predictive mode is respectively: 7,14,6,13,1,12,5,11,4,15,8,16,2,17,9,18,10.Piece to 4 × 4 carries out the pattern that pattern sub-sampling only retains half, i.e. preference pattern 7,6,1,5,4,8,2,9,10.

Canny operator edge detection step is as follows: first apply Gaussian filter smoothed image, ask amplitude and the direction of finite difference mean value computation shade of gray in 3 × 3 neighborhoods.

Further, in described step 2, adopt morphologic method to carry out refinement to the thick edge of the width detecting, that is, carry out morphologic erosion operation, the template elements of its corrosion is

$\left[\begin{array}{ccc}0& 1& 0\\ 1& 1& 1\\ 0& 1& 0\end{array}\right]$

Further, in described step 3, adopt threshold value T _hcarry out marginal point extraction.Adopt iterative method to choose optimal threshold.Iterative method performing step is as follows:

Obtain minimum value Z in the current PU of image _min, maximum Z _max, Initial Hurdle T ₀=(Z _min+ Z _max)/2.

According to threshold values T _kimage is divided into fringe region and non-fringe region by (k is iterations), calculates two-part mean value Z _o, Z _b,

$Z_O=\underset{i=0}{\overset{T_k}{\mathrm{\Σ}}}{h}_i\·i/\underset{i=0}{\overset{T_k}{\mathrm{\Σ}}}{h}_i$

$Z_B=\underset{j=T_k+1}{\overset{255}{\mathrm{\Σ}}}{h}_j\·j/\underset{j=T_k+1}{\overset{255}{\mathrm{\Σ}}}{h}_j$

Here h, _iand h _jrepresent respectively the value of the pixel of edge and non-fringe region.Calculate Z _o, Z _bafter, use formula

T _k+1=(Z _O+Z _B)/2

Calculate the threshold values T that makes new advances _k+1.

If T _k+1=T _k, i.e. T _kfor required threshold value, algorithm leaves it at that, otherwise forwards step (2) to.Iterative computation is until converge on certain stable threshold value, and this threshold value is final result T _h.

Here adopt dual threshold method from candidate marginal, detect and be connected final edge.In the time of the extraction of marginal point, the threshold value adopting is T _h, the threshold value adopting when the connection of marginal point should compare T _hlittle, to connect edge.The threshold value adopting in the time of the connection of marginal point in the present invention is T _h/ 2.,, after the first and second steps are processed, value is greater than T _hpixel be marginal point.Then taking these marginal points as seed, find the value being adjacent and be greater than T _h/ 2 pixel, adds as edge pixel point.Then, repeat this process, until all pixels all scan and go in PU.

Pre-measuring angle, border angular range and the tangent scope of each predictive mode of table 18 × 8,16 × 16,32 × 32 size PU

Pre-measuring angle, border angular range and the tangent scope of each predictive mode of table 2 64 × 64 size PU

Predictive mode	Pre-measuring angle	Border angular range	Tangent value scope
				1	-90	(-45，-135)	(-∞，-1)∪(1，+∞)
2	0	[-45，45]	[-1，1]

Further, in step 6, the scope of statistics of the tangent value to linear edge angle as shown in Table 1 and Table 2.For example, as shown in table 1, for candidate pattern 7, the scope of statistics of its angle tangent value be (0.7417 ,-1.1033]., the straight line pixel point value within the scope of initial this sets to 0, C ₇=0, if then there is the tangent value of the angle of linear edge to drop on this scope, C ₇=C ₇+ a _i, wherein a _ithe all pixel numbers in current PU on this straight line.All candidate pattern are all carried out to this and process, obtained the statistic histogram of desired each pattern.Because pattern 0, mode 3 are non-directivity predictive modes, therefore do not added up in table 1, table 2, they are all used as candidate pattern and enter into roughly selecting of RDO.

In the each PU size of table 3HM4.0, enter into the number of the candidate's of RMD rougher process predictive mode

4×4	19
		8×8	35
16×16	35
		32×32	35
64×64	4

In the each PU size of table 4 the present invention, enter into the number of the candidate's of RMD rougher process predictive mode

4×4	9
		8×8	7
16×16	5
		32×32	3
64×64	1

Further, the process of step 8 is as shown in table 4.For example, for the PU of 16 × 16 sizes, 5 candidate pattern in above statistic histogram intermediate value maximum will be selected.From the contrast of table 3 and table 4, can find out that the present invention has greatly reduced the number of the candidate's who enters into RMD rougher process pattern.Like this, can reduce widely amount of calculation, accelerate the speed of service of HEVC standard.

In the each PU size of table 5HM4.0, enter into the number of the candidate's of RDO refining process predictive mode

4×4	8
		8×8	8
16×16	3
		32×32	3
64×64	3

In the each PU size of table 6 the present invention, enter into the number of the candidate's of RDO refining process predictive mode

4×4	5
		8×8	4
16×16	3
		32×32	2
64×64	1

Further, the process of step 9 is as shown in table 6.For example, for the PU of 16 × 16 sizes, 3 candidate pattern that are chosen in optimum in RMD process are entered into the selection course of final RDO.From the contrast of table 5 and table 6, can find out that the present invention has greatly reduced the number of the candidate's who enters into RDO refining process pattern.Like this, can reduce widely amount of calculation, accelerate the speed of service of HEVC standard.

In sum, the present invention proposes the new method of the infra-frame prediction in a kind of current up-to-date coding standard HEVC in the world, to improve its speed of service.The innovative point of the method is: utilize Canny operator to carry out rim detection, and utilize Hough conversion to carry out Straight edge inspection, with according to these heuristic informations, extract the most probable edge of current predicting unit trend.Then, walk according to these edges the predictive mode of always selecting and reducing the candidate who needs detection.With this, greatly reduce the computation complexity of the coding of HEVC, improve its speed of service.

Below with reference to accompanying drawing, the technique effect of design of the present invention, concrete structure and generation is described further, to understand fully object of the present invention, feature and effect.

Brief description of the drawings

Fig. 1 is the schematic diagram of the intra prediction mode in HEVC standard;

Fig. 2 is the flow chart of new method proposed by the invention;

Fig. 3 is method and code check and the distortion factor performance curve comparison diagram of original method (HM4.0) to video test sequence Party Scene (832 × 480p resolution) proposed by the invention;

Fig. 4 is method and code check and the distortion factor performance curve comparison diagram of original method (HM4.0) to video test sequence Slide Editing (720p resolution) proposed by the invention,

Fig. 5 is method and code check and the distortion factor performance curve comparison diagram of original method (HM4.0) to video test sequence Kimonol (1080p resolution) proposed by the invention;

Fig. 6 is method and code check and the distortion factor performance curve comparison diagram of original method (HM4.0) to video test sequence SteamLocomotiveTrain (4k × 2k resolution) proposed by the invention;

Embodiment

Below in conjunction with accompanying drawing, embodiments of the invention are elaborated: the present embodiment is implemented under with technical solution of the present invention prerequisite, provided detailed execution mode and concrete operating process, but protection scope of the present invention is not limited to following embodiment.

The infra-frame prediction that can find out original HEVC from accompanying drawing 1 has a lot of candidate pattern, determines that the pattern of optimizing has larger computation complexity.

The present invention utilizes C Plus Plus on the basis of the program of the HM4.0 of HEVC, utilizes OpenCV function library, and proposed method is realized and tested.

As shown in Figure 2, the method for the computation complexity of its infra-frame prediction of reduction of the present invention comprises the steps:

Step 1, to current PU, judges whether its size is 4 × 4.In this way, carry out step-length and be 1 pattern sub-sampling, and leap to step 8.Otherwise, utilize Canny operator to carry out computing to current PU.For the PU of 4 × 4 sizes, because its size is too little, the correlation of the linear edge direction therefrom extracting and the direction of predictive mode is not large especially, therefore leaps to step 8.Be greater than the PU of 4 × 4 sizes for size, the correlation of the linear edge direction therefrom extracting and the direction of predictive mode is very large, therefore can be used for determining the candidate pattern that will detect.In OpenCV, carry out rim detection by function cvCanny access Canny operator, its function call prototype is void cvCanny (const CvArr*image, CvArr*edges, double threshold1, double threshold2, int aperture_size=3).These edges that identify in the edge of function cvCanny employing Canny algorithm search input picture and output image.

Step 2, according to the result of the computing of Canny operator, carries out refinement with morphologic method edge.Use as follows

$\left[\begin{array}{ccc}0& 1& 0\\ 1& 1& 1\\ 0& 1& 0\end{array}\right]$

Template elements is greater than 3 edge to width and carries out the corrosion operation in morphology.

Step 3, selects suitable threshold value to carry out extraction and the connection of marginal point in PU.Here adopt threshold value T, _hcarry out the extraction of marginal point, and choose optimal threshold by iterative method.The performing step of iterative method is as follows:

Obtain minimum value Z in the current PU of image _min, maximum Z _max, Initial Hurdle T ₀=(Z _min+ Z _max)/2.

According to threshold values T _kimage is divided into fringe region and non-fringe region by (k is iterations), calculates two-part mean value Z _o, Z _b,

$Z_O=\underset{i=0}{\overset{T_k}{\mathrm{\Σ}}}{h}_i\·i/\underset{i=0}{\overset{T_k}{\mathrm{\Σ}}}{h}_i$

$Z_B=\underset{j=T_k+1}{\overset{255}{\mathrm{\Σ}}}{h}_j\·j/\underset{j=T_k+1}{\overset{255}{\mathrm{\Σ}}}{h}_j$

Here h, _iand h _jrepresent respectively the value of the pixel of edge and non-fringe region.Calculate Z _o, Z _bafter, use formula

T _k+1=(Z _O+Z _B)/2

Calculate the threshold values T that makes new advances _k+1.

If T _k+1=T _k, i.e. T _kfor required threshold value, algorithm leaves it at that, otherwise forwards step (2) to.Iterative computation is until converge on certain stable threshold value, and this threshold value is final result T _h.

Here adopt dual threshold method from candidate marginal, detect and be connected final edge.In the time of the extraction of marginal point, the threshold value adopting is T _h, the threshold value adopting when the connection of marginal point should compare T _hlittle, to connect edge.The threshold value adopting in the time of the connection of marginal point in the present invention is T _h/ 2.,, after the first and second steps are processed, value is greater than T _hpixel be marginal point.Then taking these marginal points as seed, find the value being adjacent and be greater than T _h/ 2 pixel, adds as edge pixel point.Then, repeat this process, until all pixels all scan and go in PU.

Step 4, to current PU, utilizes Hough change detection linear edge pixel,, retains the marginal point on a certain straight line that is, removes the marginal point on curve.In OpenCV, carry out the detection of straightway by function cvHoughLines2.The function call prototype of cvHoughLines2 is CvSeq*cvHoughLines2 (CvArr*image, void*line storage, int method, double rho, double theta, int threshold, double param1, double param2).The present invention is arranged to CV_HOUGH_PROBABILISTIC by parameter m ethod and represents to select probability Hough transformation, carries out the detection of straightway.Threshold is threshold parameter, if corresponding aggregate-value is greater than it, function can return to this line segment.Param1 represents minimum line segment length, and param2 is illustrated in the largest interval value of carrying out segment link on same straight line.

Step 5, to each linear edge, determines the differential seat angle θ in itself and horizontal direction, and its tangent value tan θ.

Step 6, according to the scope of the tangent value of the pre-measuring angle of each intra prediction mode, with the tangent value of each linear edge in the current PU detecting, determines and in current PU, drops on the linear edge pixel sum within the scope of each with statistic histogram.The scope of statistics of the tangent value to linear edge angle as shown in Table 1 and Table 2.For example, as shown in table 1, for candidate pattern 7, the scope of statistics of its angle tangent value be (0.7417 ,-1.1033]., the straight line pixel point value within the scope of initial this sets to 0, C ₇=0, if then have the tangent value of the angle of linear edge drop on this scope (0.7417 ,-1.1033] in, C ₇=C ₇+ a _i, wherein a _ithe all pixel numbers in current PU on this straight line.All candidate pattern are all carried out to this and process, obtained the statistic histogram of desired each pattern.Because pattern 0, mode 3 are non-directivity predictive modes, therefore do not added up in table 1, table 2, they are all fixed as candidate pattern and enter into roughly selecting of RDO.

Step 7, according to this statistic histogram, determines several predictive modes of the linear edge pixel sum maximum in histogram.

Step 8, according to the size of current PU, and the statistic histogram obtaining, definite candidate's who enters pattern rougher process RMD frame mode.Further, to the PU of a certain size, the number of the candidate's who enters RDO detection of its selection predictive mode is as shown in table 6.For example, for the PU of 16 × 16 sizes, 5 candidate pattern in above statistic histogram intermediate value maximum will be selected.From the contrast of table 3 and table 4, can find out that the present invention has greatly reduced the number of the candidate's who enters into RMD rougher process pattern.Like this, can reduce widely amount of calculation, accelerate the speed of service of HEVC standard.

Step 9, according to RMD process, and the size of current PU, definite candidate's who enters pattern refining process RDO frame mode.Further, to the PU of a certain size, the number of the candidate's who enters RDO detection of its selection predictive mode is as shown in table 6.For example, for the PU of 16 × 16 sizes, 3 candidate pattern that are chosen in optimum in RMD process are entered into the selection course of final RDO.From the contrast of table 5 and table 6, can find out that the present invention has greatly reduced the number of the candidate's who enters into RDO rougher process pattern.Like this, can reduce widely amount of calculation, accelerate the speed of service of HEVC standard.

Step 10, according to the result of RDO, determines the intra prediction mode of the optimization of current PU.

The present invention, according to general test environment, tests proposed method.Coded format is full I frame structure, and test QP point is 37,32,27,22, uses adaptive binary arithmetic coding CABAC (Context-based Adaptive Binary Arithmetic Coding) to carry out entropy coding to video sequence.Consider that HEVC can be widely applied in high-resolution video sequence future, test mainly, taking high definition and SD cycle tests as main, mainly contains 25601 × 600p, 1920 × 1080p, 1280 × 720p and 832 × 480p, the video test sequence of totally 4 kinds of sizes.Here the minimizing of the scramble time mainly bringing from proposed new method and the corresponding cost of paying are considered the performance of the method.Performance index are mainly expressed the performance of institute's put forward the methods particularly by parameters such as BD-PSNR/Rate, Δ B, Δ P and Δ T.

The experimental results of table 7 method proposed by the invention

Experimental result is as shown in table 7, if positive number in all data results represents that under this index, method of the present invention increases with respect to its numerical value of former method, if negative represent under this index, method of the present invention reduces with respect to its numerical value of former method.The index of test performance is defined as follows:

(1) the scramble time Δ T saving,

$\mathrm{\ΔT}=\frac{T_{\mathrm{proposed}}-T_{\mathrm{HM}4.0}}{T_{\mathrm{HM}4.0}}$

Wherein, T _hM4.0represent to use the scramble time of the existing algorithm of HM4.0, T _proposedit is the scramble time of the algorithm of carrying in literary composition.Obtain under equal code efficiency condition Δ T larger, presentation code end computation complexity reduces more, and the performance of the method proposing is better.

(2) method proposing and HM4.0 intraframe prediction algorithm code check difference DELTA B,

$\mathrm{\ΔB}=\frac{B_{\mathrm{Proposed}}-B_{\mathrm{HM}4.0}}{B_{\mathrm{HM}4.0}}$

Wherein, B _hM4.0represent the encoder bit rate of existing HM4.0 method, B _proposedit is the encoder bit rate of method proposed by the invention.Δ B is less, represents that the method and the existing methodical coding efficiency of HM4.0 that propose are more approaching.

(3) the difference DELTA P of the Y-PSNR (Peak Signal to Noise Ratio, PSNR) of the method proposing and HM4.0 intra-frame prediction method luminance component,

ΔP=P _Proposed-P _HM4.0

Wherein, P _hM4.0represent the Y-PSNR PSNR of the existing methodical luminance component of HM4.0, P _proposedthe PSNR of the luminance component of method proposed by the invention.Δ P is less, represents that the method and the existing methodical coding efficiency of HM4.0 that propose are more approaching.

BD-Rate represents under same Y-PSNR (PSNR), the added value of code check.BD-PSNR represents under same code check, the minimizing value of PSNR.

From experimental result table 7, can find out, the method proposing only has the reduction of minimum Y-PSNR (PSNR) and minimum code check to increase, and therefore it has good performance.Meanwhile, table can be found out thus, and method proposed by the invention can be saved minimum 16.15%, maximum with respect to the standard method in HM4.0 can reach for 32.73% scramble time.

The code check to the video test sequence under different resolution to institute's put forward the methods and former method (HM4.0) being obtained by experimental result and the curve of the distortion factor (Rate-Distortion) are as shown in Fig. 3,4,5,6.The abscissa of these figure is code check, and unit is thousand bit numbers (kbps) that need transmission per second.The ordinate of these figure is Y-PSNR (PSNR), and unit is dB.From these figure, can find out that the code check of method proposed by the invention and former method and two curves of the distortion factor are substantially the same.Thus, can illustrate that proposed method has kept the precision of predictive mode, under the condition of identical code check, not worsen or reduce than significantly the distortion factor.

From table 7 and Fig. 3,4,5,6, can find out, the method proposing, with respect to original method, can on average save for 23% scramble time.Meanwhile, the fluctuation of its code check and PSNR is less, and for the calculation cost of saving, the reduction in performance almost can be ignored.Visible, method proposed by the invention has good performance.

More than describe preferred embodiment of the present invention in detail.The ordinary skill that should be appreciated that this area just can design according to the present invention be made many modifications and variations without creative work.Therefore, all technical staff in the art, all should be in by the determined protection range of claims under this invention's idea on the basis of existing technology by the available technical scheme of logical analysis, reasoning, or a limited experiment.

Claims (7)

1. In the international up-to-date video compression standard HEVC of 1.Yi Zhong, determine rapidly the method for intraframe predictive coding pattern.Described method comprises:

   Step 1, to current predicting unit (PU), judges whether its size is 4 × 4.In this way, carry out step-length and be 1 pattern sub-sampling, and leap to step 8.Otherwise, utilize Canny operator to carry out computing to current PU.

   Step 2, according to the result of the computing of Canny operator, carries out refinement with morphologic method edge.

   Step 3, selects suitable threshold value to carry out extraction and the connection of marginal point in PU.

   Step 4, to current PU, utilizes Hough change detection linear edge pixel,, retains the marginal point on a certain straight line that is, removes the marginal point on curve.

   Step 5, to each linear edge, determines the differential seat angle θ in itself and horizontal direction, and its tangent value tan θ.

   Step 6, according to the scope of the tangent value of the pre-measuring angle of each intra prediction mode, with the tangent value of each linear edge in the current PU detecting, determines and in current PU, drops on the linear edge pixel sum within the scope of each with statistic histogram.

   Step 7, according to this statistic histogram, determines several predictive modes of the linear edge pixel sum maximum in histogram.

   Step 8, according to the size of current PU, and the statistic histogram obtaining, definite candidate's who enters pattern rougher process (RMD) frame mode.

   Step 9, according to RMD process, and the size of current PU, definite candidate's who enters pattern refining process (RDO) frame mode.

   Step 10, according to the result of RDO, determines the intra prediction mode of the optimization of current PU.
2. 2. the method for the refinement at a kind of edge that width is greater than to 3 as claimed in claim 1.In described step 2, adopt morphologic method to be greater than 3 edge to the width detecting and carry out refinement, that is, carry out morphologic erosion operation, the template elements of its corrosion is

   

   。
3. 3. a kind of method that adopts iterative method to choose optimal threshold extraction marginal point as claimed in claim 1.In described step 3, adopt threshold value T _hcarry out marginal point extraction.Adopt iterative method to choose optimal threshold.Iterative method performing step is as follows:

   Obtain minimum value Z in the current PU of image _min, maximum Z _max, Initial Hurdle T ₀=(Z _min+ Z _max)/2.According to threshold values T _kimage is divided into edge pixel point and non-edge pixel point by (k is iterations), calculates two-part mean value Z _o, Z _b,

   

   

   Here h, _iand h _jrepresent respectively the value of edge pixel point and non-edge pixel point.Calculate Z _o, Z _bafter, use formula

   T _k+1=(Z _O+Z _B)/2

   Calculate the threshold values T that makes new advances _k+1.

   If T _k+1=T _k, i.e. T _kfor required threshold value, algorithm leaves it at that, otherwise forwards step (2) to.Iterative computation is until converge on certain stable threshold value, and this threshold value is final result T _h.
4. 4. employing dual threshold method as claimed in claim 1., from candidate marginal, detect and the method that is connected final edge pixel.In described step 3, in the time of the extraction of marginal point, the threshold value adopting is T _h, the threshold value adopting when the connection of marginal point should compare T _hlittle, to connect edge.The threshold value adopting in the time of the connection of marginal point in the present invention is T _h/ 2.,, after the first and second steps are processed, value is greater than T _hpixel be marginal point.Then taking these marginal points as seed, find the value being adjacent and be greater than T _h/ 2 pixel, adds as edge pixel point.Then, repeat this process, until all pixels all scan and go in PU.
5. 5. the method for the edge trend of determining current PU as claimed in claim 1., according to the scope of the tangent value of the pre-measuring angle of each intra prediction mode, with the tangent value of each linear edge in the current PU detecting, determine and in current PU, drop on the linear edge pixel sum within the scope of each with statistic histogram.
6. 6. as claimed in claim 1, according to the size of current PU, and the statistic histogram obtaining, the method for definite candidate pattern that enters RMD testing process.To size 4 × 4,8 × 8,16 × 16,32 × 32 and 64 × 64, select respectively 9,7,5,3,1 candidate pattern to enter into the testing process of RMD.Wherein PU is of a size of 8 × 8,16 × 16,32 × 32 and 64 × 64 predicting unit, selects the large candidate pattern of statistic histogram cathetus edge pixel point sum to enter into RMD testing process.
7. 7. as claimed in claim 1, according to the size of current PU, and the result of RMD, the method for definite candidate pattern that enters RDO testing process.To size 4 × 4,8 × 8,16 × 16,32 × 32 and 64 × 64, select respectively 5,4,3,2,1 candidate pattern to enter into the testing process of RDO.After the standard of selecting is RMD process, code check and the little candidate pattern of predicted distortion degree enter into the testing process of RDO.

Images