CN101895755A - Quick 4*4 block intra-frame prediction mode selecting method - Google Patents

Abstract

The invention relates to a quick 4*4 block intra-frame prediction mode selecting method. The method comprises the following steps of: dividing a picture into 4*4 block images by using an H.264 coder frame, and calculating seven edge directional coefficients of each 4*4 block; dividing each 4*4 image block into nine edge modes comprising one consistent regional block edge mode and eight different directional edge modes according to the edge directional coefficients; then selecting one or three intra-frame prediction modes according to the edge modes of the image block, and if the three prediction modes are selected, determining a final intra-frame prediction mode according to the calculated rate distortion cost value after pre-coding; and if only one prediction mode is selected, taking the prediction mode as the final prediction mode. The method quickly determines the 4*4 block intra-frame prediction mode of the H.264 under the condition of low calculation complexity, and ensures good rate distortion efficiency.

Description

A kind of quick 4x4 block selection method of intra-frame prediction mode

Technical field

The invention belongs to the digital video coding field, relate to a kind of based on the quick 4x4 block selection method of intra-frame prediction mode of encoder H.264.

Background technology

Up-to-date video compression standard is H.264 in the encoder, in the frame and interframe encode has adopted rich and varied coding mode option.In the intraframe predictive coding pattern, H.264 encoder is supported 4 kinds of Intra_16x16 predictive modes and 9 kinds of Intra_4x4 predictive modes to luminance component, to 4 kinds of Intra_8x8 predictive modes of chrominance block support such as Intra_16x16.In 9 kinds of predictive modes for the 4x4 fritter, Intra_4x4_DC pattern wherein is the mean value that utilizes some the decoded area pixel adjacent with this 4x4 fritter, and this fritter is made a prediction; Corresponding respectively 8 the different orientation angles of other 8 kinds of predictive modes are predicted.In video coding process H.264, select a kind of suitable predictive mode process to be called model selection in the odd encoder pattern of comforming.These abundant coding modes have greatly improved compression efficiency.But the complexity of model selection process also increases greatly.In the identifying code of encoder H.264, adopted the RDO frame model of optimizing based on Lagrange, and adopted full way of search to seek optimum prediction coding mode on a kind of rate distortion meaning.This all direction search method need carry out precoding, calculation rate distortion cost value to all predictive modes, and the pattern of last selection rate distortion cost value minimum is final predictive coding pattern.The computation complexity of this method is that real-time encoder institute is unacceptable.So, need the model selection of encoder be optimized the not loss coding compression efficiency of trying one's best simultaneously.

Summary of the invention

The purpose of this invention is to provide a kind of based on the quick 4x4 block selection method of intra-frame prediction mode of encoder H.264.

The inventive method comprises the steps:

Step (1) is divided into the piece image of 4x4 size to a width of cloth picture according to encoder frame H.264, calculates seven edge direction coefficients of each 4x4 image block; According to the edge direction coefficient each 4x4 image block is divided into nine kinds of edge pattern: comprise a kind of consistency region unit edge pattern and eight different directions edge pattern; Seven edge direction coefficient methods that calculate each 4x4 image block are as follows:

Wherein H, V, X, A, B, C, D represent seven edge direction coefficients, S respectively ₀₀, S ₀₁, S ₁₀, S ₁₁Represent upper left, upper right, the lower-left of a piece, the mean value of the four/pixels in bottom right respectively,

U, v=0,1, i, j are the index value of four/pixels;

Wherein p (x y) is the gray value of pixel, x, y=0,1,2,3 be vertical direction from top to bottom with horizontal direction index value from left to right; Scale1=QPstep/16, wherein QPstep is a quantization step, scale2 is that number of pixels is divided by 2;

Smallest positive integral is got in expression; || absolute value is asked in expression.

According to above seven direction coefficients, the method that each 4x4 image block is divided into nine kinds of edge pattern is as follows:

If (1-1) H=V=0, then this 4x4 image block belongs to consistency region unit edge pattern, the direct current predictive mode of the corresponding 4x4 infra-frame prediction of this edge pattern, and promptly intra prediction mode 2;

(1-2) existing | under the situation of H|=|V| ≠ 0, if Abs (A-D)＞Abs (C-B), then this piece belongs to the DL edge pattern, and the edge direction of this edge pattern correspondence is consistent with the prediction direction of 4x4 piece intra prediction mode 3; Otherwise this piece belongs to the DR edge pattern, and the edge direction of this edge pattern correspondence is consistent with the prediction direction of 4x4 intra prediction mode 4;

(1-3) exist | H|＞| under the situation of V|: during V=0, then this piece belongs to the H edge pattern, and the prediction direction that the edge direction of this edge pattern correspondence and intra prediction mode 1 are corresponding is consistent; V ≠ 0 o'clock, if A or D are the maximums in A, B, C, D four numbers, then this piece belongs to the HU edge pattern, the edge direction of this edge pattern correspondence is consistent with intra prediction mode 8 corresponding prediction direction; If B or C are the maximums in A, B, C, D four numbers, then this piece belongs to the HD edge pattern, and the prediction direction that the edge direction of this edge pattern correspondence and intra prediction mode 6 are corresponding is consistent;

(1-4) exist | H|＜| under the situation of V|: during H=0, then this piece belongs to the V edge pattern, and the prediction direction that the edge direction of this edge pattern correspondence and intra prediction mode 0 are corresponding is consistent; H ≠ 0 o'clock, if A or D are the maximums in A, B, C, D four numbers, then this piece belongs to the VL edge pattern, the edge direction of this edge pattern correspondence is consistent with intra prediction mode 7 corresponding prediction direction; If B or C are the maximums in A, B, C, D four numbers, then this piece belongs to the VR edge pattern, and the prediction direction that the edge direction of this edge pattern correspondence and intra prediction mode 5 are corresponding is consistent.

Intra prediction mode in the inventive method numbering is numbered consistent with the pattern of encoder defined H.264.

Step (2) is selected a kind of or three kinds of intra prediction modes according to the edge pattern of image block, if select three kinds of predictive modes, then determines last intra prediction mode according to the rate distortion costs value that calculates after the precoding; And for only selecting a kind of predictive mode, this predictive mode is exactly last predictive mode.

The direction of described edge pattern correspondence according to image block is selected a kind of prediction mode method: if this piece belongs to consistency region unit pattern, directly select the last coding mode of DC predictive mode for this 4x4 piece; If this piece does not belong to the consistency region unit, directly selecting the predictive mode consistent with the edge direction of this piece is the last predictive mode of this piece;

The direction of described edge pattern correspondence according to image block is selected three kinds of prediction mode method: if this piece belongs to consistency region unit pattern, directly select the last coding mode of DC predictive mode for this 4x4 piece; If this piece does not belong to the consistency region unit, three kinds of predictive modes selecting are that the pairing predictive mode of this block edge direction adds two predictive modes that this direction is the most contiguous, to calculating the rate distortion costs value after these three predictive mode precodings, rate distortion costs value minimum be defined as last predictive mode; Calculation rate distortion cost value J (s, c, MODE|QP, λ _MODE) method as follows:

J(s，c，MODE|QP，λ _MODE)＝SSD(s，c，MODE|QP)+λ _MODE×R(s，c，MODE|QP)

SSD (s, c MODE|QP) are the quadratic sum of the difference of current block and reconstructed block:

$\mathrm{SSD}=(s,c,\mathrm{MODE}|\mathrm{QP})=\underset{x=0}{\overset{B_1}{\mathrm{\Σ}}}\underset{y=0}{\overset{B_2}{\mathrm{\Σ}}}{\left|s\right[x,y]-c(x,y)|}^2$

B ₁Horizontal number of pixels, the B of expression piece ₂The Vertical number of pixels of expression piece gets 4, and s (x, y) and be respectively original image and reconstructed image pixel value; (MODE|QP) code check behind the expression entropy coding comprises coded prediction pattern, DCT coefficient to R for s, c; λ _MODE=0.85 * 2 ^QP/3Be Lagrange multiplier, QP is a quantization parameter.

The inventive method selects a kind of reasonable 4x4 intra prediction mode to encode according to the edge pattern of each 4x4 piece fast, under situation about guaranteeing than high coding efficiency, saves encoder complexity in a large number.Compare with other fast algorithm, the needed extra computation complexity of this method is almost nil, is fit to very much be applied to the especially limited video coding occasion of computational resource of video encoder.The inventive method determine 4x4 piece intra prediction mode H.264 fast, and fraction distortion efficient is good in the very low situation of computation complexity.

Description of drawings

Fig. 1 is the prediction direction schematic diagram of nine kinds of different predictive mode correspondences among the present invention.

Embodiment

A kind of quick 4x4 block selection method of intra-frame prediction mode, concrete steps are as follows:

Step (1) is divided into the piece image of 4x4 size to a width of cloth picture according to encoder frame H.264, calculates seven edge direction coefficients of each 4x4 image block; According to the edge direction coefficient each 4x4 image block is divided into nine kinds of edge pattern: comprise a kind of consistency region unit edge pattern and eight different directions edge pattern; Seven edge direction coefficient methods that calculate each 4x4 image block are as follows:

Wherein H, V, X, A, B, C, D represent seven edge direction coefficients, S respectively ₀₀, S ₀₁, S ₁₀, S ₁₁Represent upper left, upper right, the lower-left of a piece, the mean value of the four/pixels in bottom right respectively,

U, v=0,1, i, j are the index value of four/pixels;

Wherein p (x y) is the gray value of pixel, x, y=0,1,2,3 be vertical direction from top to bottom with horizontal direction index value from left to right; Scale1=QPstep/16, wherein QPstep is a quantization step, scale2 is that number of pixels is divided by 2; Smallest positive integral is got in expression.

According to above seven direction coefficients, the method that each 4x4 image block is divided into nine kinds of edge pattern is as follows:

If (1-1) H=V=0, then this 4x4 image block belongs to consistency region unit edge pattern, the direct current predictive mode of the corresponding 4x4 infra-frame prediction of this edge pattern, and promptly intra prediction mode 2;

(1-2) existing | under the situation of H|=|V| ≠ 0, if Abs (A-D)＞Abs (C-B), then this piece belongs to the DL edge pattern, and the edge direction of this edge pattern correspondence is consistent with the prediction direction of 4x4 piece intra prediction mode 3; Otherwise this piece belongs to the DR edge pattern, and the edge direction of this edge pattern correspondence is consistent with the prediction direction of 4x4 intra prediction mode 4;

(1-3) exist | H|＞| under the situation of V|: during V=0, then this piece belongs to the H edge pattern, and the prediction direction that the edge direction of this edge pattern correspondence and intra prediction mode 1 are corresponding is consistent; V ≠ 0 o'clock, if A or D are the maximums in A, B, C, D four numbers, then this piece belongs to the HU edge pattern, the edge direction of this edge pattern correspondence is consistent with intra prediction mode 8 corresponding prediction direction; If B or C are the maximums in A, B, C, D four numbers, then this piece belongs to the HD edge pattern, and the prediction direction that the edge direction of this edge pattern correspondence and intra prediction mode 6 are corresponding is consistent;

(1-4) exist | H|＜| under the situation of V|: during H=0, then this piece belongs to the V edge pattern, and the prediction direction that the edge direction of this edge pattern correspondence and intra prediction mode 0 are corresponding is consistent; H ≠ 0 o'clock, if A or D are the maximums in A, B, C, D four numbers, then this piece belongs to the VL edge pattern, the edge direction of this edge pattern correspondence is consistent with intra prediction mode 7 corresponding prediction direction; If B or C are the maximums in A, B, C, D four numbers, then this piece belongs to the VR edge pattern, and the prediction direction that the edge direction of this edge pattern correspondence and intra prediction mode 5 are corresponding is consistent.

It is consistent that intra prediction mode in this method numbering and the pattern of encoder defined are H.264 numbered, as shown in Figure 1.Numeral is the pattern numbering of different predictive modes among the figure.

Step (2) is selected a kind of or three kinds of intra prediction modes according to the edge pattern of image block, if select three kinds of predictive modes, then determines last intra prediction mode according to the rate distortion costs value that calculates after the precoding; And for only selecting a kind of predictive mode, this predictive mode is exactly last predictive mode.

Direction according to the edge pattern correspondence of image block is selected a kind of prediction mode method: if this piece belongs to consistency region unit pattern, directly select the last coding mode of DC predictive mode for this 4x4 piece; If this piece does not belong to the consistency region unit, directly selecting the predictive mode consistent with the edge direction of this piece is the last predictive mode of this piece;

Direction according to the edge pattern correspondence of image block is selected three kinds of prediction mode method: if this piece belongs to consistency region unit pattern, directly select the last coding mode of DC predictive mode for this 4x4 piece; If this piece does not belong to the consistency region unit, three kinds of predictive modes selecting are that the pairing predictive mode of this block edge direction adds two predictive modes that this direction is the most contiguous, to calculating the rate distortion costs value after these three predictive mode precodings, rate distortion costs value minimum be defined as last predictive mode; Calculation rate distortion cost value J (s, c, MODE|QP, λ _MODE) method as follows:

J(s，c，MODE|QP，λ _MODE)＝SSD(s，c，MODE|QP)+λ _MODE×R(s，c，MODE|QP)

SSD (s, c MODE|QP) are the quadratic sum of the difference of current block and reconstructed block:

$\mathrm{SSD}=(s,c,\mathrm{MODE}|\mathrm{QP})=\underset{x=0}{\overset{B_1}{\mathrm{\Σ}}}\underset{y=0}{\overset{B_2}{\mathrm{\Σ}}}{\left|s\right[x,y]-c(x,y)|}^2$

B ₁Horizontal number of pixels, the B of expression piece ₂The Vertical number of pixels of expression piece gets 4, and s (x, y) and be respectively original image and reconstructed image pixel value; (MODE|QP) code check behind the expression entropy coding comprises coded prediction pattern, DCT coefficient to R for s, c; λ _MODE=0.85 * 2 ^QP/3Be Lagrange multiplier, QP is a quantization parameter.

Claims (1)

1. a quick 4x4 block selection method of intra-frame prediction mode is characterized in that this method comprises the steps:

Step (1) is divided into the piece image of 4x4 size to a width of cloth picture according to encoder frame H.264, calculates seven edge direction coefficients of each 4x4 image block; According to the edge direction coefficient each 4x4 image block is divided into nine kinds of edge pattern: comprise a kind of consistency region unit edge pattern and eight different directions edge pattern; Seven edge direction coefficient methods that calculate each 4x4 image block are as follows:

Wherein H, V, X, A, B, C, D represent seven edge direction coefficients, S respectively ₀₀, S ₀₁, S ₁₀, S ₁₁Represent upper left, upper right, the lower-left of a piece, the mean value of the four/pixels in bottom right respectively,

U, v=0,1, i, j are the index value of four/pixels;

Wherein p (x y) is the gray value of pixel, x, y=0,1,2,3 be vertical direction from top to bottom with horizontal direction index value from left to right; Scale1=QPstep/16, wherein QPstep is a quantization step, scale2 is that number of pixels is divided by 2;

Smallest positive integral is got in expression; || absolute value is asked in expression;

According to above seven direction coefficients, the method that each 4x4 image block is divided into nine kinds of edge pattern is as follows:

If (1-1) H=V=0, then this 4x4 image block belongs to consistency region unit edge pattern, the direct current predictive mode of the corresponding 4x4 infra-frame prediction of this edge pattern, and promptly intra prediction mode 2;

(1-2) existing | under the situation of H|=|V| ≠ 0, if Abs (A-D)＞Abs (C-B), then this piece belongs to the DL edge pattern, and the edge direction of this edge pattern correspondence is consistent with the prediction direction of 4x4 piece intra prediction mode 3; Otherwise this piece belongs to the DR edge pattern, and the edge direction of this edge pattern correspondence is consistent with the prediction direction of 4x4 intra prediction mode 4;

(1-3) exist | H|＞| under the situation of V|: during V=0, then this piece belongs to the H edge pattern, and the prediction direction that the edge direction of this edge pattern correspondence and intra prediction mode 1 are corresponding is consistent; V ≠ 0 o'clock, if A or D are the maximums in A, B, C, D four numbers, then this piece belongs to the HU edge pattern, the edge direction of this edge pattern correspondence is consistent with intra prediction mode 8 corresponding prediction direction; If B or C are the maximums in A, B, C, D four numbers, then this piece belongs to the HD edge pattern, and the prediction direction that the edge direction of this edge pattern correspondence and intra prediction mode 6 are corresponding is consistent;

(1-4) exist | H|＜| under the situation of V|: during H=0, then this piece belongs to the V edge pattern, and the prediction direction that the edge direction of this edge pattern correspondence and intra prediction mode 0 are corresponding is consistent; H ≠ 0 o'clock, if A or D are the maximums in A, B, C, D four numbers, then this piece belongs to the VL edge pattern, the edge direction of this edge pattern correspondence is consistent with intra prediction mode 7 corresponding prediction direction; If B or C are the maximums in A, B, C, D four numbers, then this piece belongs to the VR edge pattern, and the prediction direction that the edge direction of this edge pattern correspondence and intra prediction mode 5 are corresponding is consistent;

Wherein intra prediction mode numbering is numbered consistent with the pattern of encoder defined H.264;

Step (2) is selected a kind of or three kinds of intra prediction modes according to the edge pattern of image block, if select three kinds of predictive modes, then determines last intra prediction mode according to the rate distortion costs value that calculates after the precoding; And for only selecting a kind of predictive mode, this predictive mode is exactly last predictive mode;

The direction of described edge pattern correspondence according to image block is selected a kind of prediction mode method: if this piece belongs to consistency region unit pattern, directly select the last coding mode of DC predictive mode for this 4x4 piece; If this piece does not belong to the consistency region unit, directly selecting the predictive mode consistent with the edge direction of this piece is the last predictive mode of this piece;

The direction of described edge pattern correspondence according to image block is selected three kinds of prediction mode method: if this piece belongs to consistency region unit pattern, directly select the last coding mode of DC predictive mode for this 4x4 piece; If this piece does not belong to the consistency region unit, three kinds of predictive modes selecting are that the pairing predictive mode of this block edge direction adds two predictive modes that this direction is the most contiguous, to calculating the rate distortion costs value after these three predictive mode precodings, rate distortion costs value minimum be defined as last predictive mode; Calculation rate distortion cost value J (s, c, MODE|QP, λ _MODE) method as follows:

J(s，c，MODE|QP，λ _MODE)＝SSD(s，c，MODE|QP)+λ _MODE×R(s，c，MODE|QP)

SSD (s, c MODE|QP) are the quadratic sum of the difference of current block and reconstructed block:

$\mathrm{SSD}=(s,c,\mathrm{MODE}|\mathrm{QP})=\underset{x=0}{\overset{B_1}{\mathrm{\Σ}}}\underset{y=0}{\overset{B_2}{\mathrm{\Σ}}}{\left|s\right[x,y]-c(x,y)|}^2$

B ₁Horizontal number of pixels, the B of expression piece ₂The Vertical number of pixels of expression piece gets 4, and s (x, y) and be respectively original image and reconstructed image pixel value; (MODE|QP) code check behind the expression entropy coding comprises coded prediction pattern, DCT coefficient to R for s, c; λ _MODE=0.85 * 2 ^QP/3Be Lagrange multiplier, QP is a quantization parameter.

Images