CN100551075C - Low-complexity in-frame estimation mode selection method - Google Patents

Low-complexity in-frame estimation mode selection method Download PDF

Info

Publication number
CN100551075C
CN100551075C CN 200710175906 CN200710175906A CN100551075C CN 100551075 C CN100551075 C CN 100551075C CN 200710175906 CN200710175906 CN 200710175906 CN 200710175906 A CN200710175906 A CN 200710175906A CN 100551075 C CN100551075 C CN 100551075C
Authority
CN
China
Prior art keywords
prediction mode
mode
step
current block
block
Prior art date
Application number
CN 200710175906
Other languages
Chinese (zh)
Other versions
CN101175212A (en
Inventor
雯 左
梁立伟
宁 王
Original Assignee
中兴通讯股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 中兴通讯股份有限公司 filed Critical 中兴通讯股份有限公司
Priority to CN 200710175906 priority Critical patent/CN100551075C/en
Publication of CN101175212A publication Critical patent/CN101175212A/en
Application granted granted Critical
Publication of CN100551075C publication Critical patent/CN100551075C/en

Links

Abstract

本发明公开了一种低复杂度的帧内预测模式选择方法,在当前块中取表征纹理特征的两组样本点,根据相邻块的开销值,计算阈值,该方法包括以下步骤:判断当前块是否处于图像帧的特殊位置,如果是,计算该特殊位置下的最优预测模式并输出,结束计算;如果不是,计算当前块在最可能预测模式下的开销值,如果小于阈值,则输出该模式,结束计算;反之,选择所有非最可能的预测模式中开销值最小的一个,作为临时最优预测模式;比较最可能预测模式与临时最优预测模式的开销值,选择其中较小的作为最优预测模式输出。 The present invention discloses an intra-forecast mode selecting method of low complexity, two samples taken characterizing texture feature points in the current block, the cost value of adjacent blocks, calculate the threshold value, the method comprising the steps of: determining the current whether the block is the special position of the image frame, and if so, computing the optimal prediction mode at that particular position, and outputs the calculation is finished; if not, calculating the cost value of a current block in the most likely prediction mode, if less than the threshold, the output this model, the calculation is finished; otherwise, all non-selected prediction mode most likely a smallest cost value as the temporary optimal prediction mode; Comparative most likely prediction mode and the optimum prediction mode temporary cost value, selecting the smaller of output as the optimal prediction mode. 本发明通过选取特征点代替整个数据块,在对视频编码质量没有明显影响的情况下,显著降低帧内预测模式选择的计算量,提高视频编码的实时性。 The present invention replace the entire block of data by selecting feature points, without significant effect on the quality of video coding, the significant reduction amount calculating an intra prediction mode selection, improve real-time video encoding.

Description

一种低复杂度的帧内预测模式选择方法 A low complexity intra prediction mode selection method

技术领域 FIELD

本发明属于视频信息压缩领域,具体涉及一种低复杂度的帧内预测模式选择方法。 The present invention belongs to the field of video information compression, particularly relates to an intra prediction mode selection A low complexity method.

背景技术 Background technique

目前的高级视频编码标准中都包含了帧内预测功能,利用临近块的样本点作外推来实现对当前块的预测,以更好的消除单帧图像内的空间冗余,这样只需要对预测块和当前块的残差进行编码。 The current advanced video coding standards are included in the intra prediction, the use of adjacent sample points for extrapolation block prediction of the current block is achieved, in order to better eliminate redundant space within the single image, so that only for residual of the current block and the prediction block is encoded. 尤其是在变化平坦的区域,利用帧内预测可以大大的降低码率。 In particular, changes in the flat region, can greatly reduce the bit rate using the intra prediction.

当一个宏块是采用帧内模式编码时,利用先前已经编码并重构的块构造一个预测块P。 When a macroblock is intra-frame coding mode, using the previously encoded and reconstructed blocks have been configured a prediction block P. 对于亮度分量,可以为每个块或者宏块创建预测块P。 For the luminance component, may each block or macro block to create a prediction block P. 例如 E.g

H.264编码标准中4x4亮度块共有9种可选才莫式,16x16亮度块有4种可选模式。 H.264 coding standard 4x4 luma block has 9 Mohs only optional, 16x16 luma block has four selectable modes.

如图l所示,在H.264标准中,利用相邻块中已经解码的13个样本点(A〜L和Q)中的几个或者所有的点,来预测当前4x4亮度块中的样本点(a〜p)。 Shown in Figure l, in the H.264 standard, the use of 13 sample points in the block has been decoded (A~L and Q) of several or all of the adjacent points, samples of the current 4x4 luminance predicted block point (a~p). 选择9种预测模式中效果最好的一种,作为该块的最佳预测模式。 Selecting the best one 9 prediction modes in effect, as the optimum prediction mode of the block.

9种预测模式包括:模式2的均值预测(DC—PRED )和如图2所示的8种方 9 prediction modes comprising: a mode shown in FIG mean prediction (DC-PRED) 2 in FIG. 2 and 8 of ways where

向预测。 To predict.

对每一个4x4块的预测模式都进行编码需要占用很多比特,适当的利用空间相邻块的相关性,可以达到高效编码的目的。 For each 4x4 block prediction modes need to take a lot of bits for encoding, by using appropriate spatial correlation of neighboring blocks, you can achieve high-efficiency encoding. 如图3所示,C是当前的4x4亮度块,根据A块和B块预测模式的不同组合得到C块的一个最可能的预测模式。 As shown in FIG 3, C is the current 4x4 luminance block, to obtain a most probable mode of C-block according to the prediction blocks of different combinations of A and B block prediction mode. 将得到的最可能的预测模式与前述C块的最佳预测模式相比较,如果相同,则只需在编码时使用1个比特表示最可能的预测模式,否则只需发送3个比特表示剩余8种预测模式中最佳的一个。 The most likely prediction mode and the optimum prediction mode obtained in block C is compared, if the same, then just use one bit represents the most likely prediction mode coding, or simply send the remaining three bits represent 8 the best prediction modes in one.

在传统的做法中,使用9种模式的全搜索方法来找到最优的一种预测模 In the traditional approach, a full nine modes search method to find the optimal method of predicting die

式,主要的步骤如下所示: Type, the main steps are as follows:

1、根据一种模式构造出4x4的预测块P; 1, constructed according to one mode of the prediction block P 4x4;

2 、计算原始块与预测块P之间的绝对误差和&4Z)16; 2, calculating the absolute error between the original block and the predicted block P and & 4Z) 16;

3 、 i十算开名肖{直CoW16 = &4£>16 + 4/U(g尸) (1 ) 3, i {ten Shore operator name apart linear CoW16 = & 4 £> 16 + 4 / U (g corpse) (1)

雄16=丄^>(义,力—W(x,力l (2) Shang male = 16 ^> (sense, force -W (x, force l (2)

其中,义(2巧是量化因子e?的指数函数,2尸根据不同的标准,取值不同; Wherein Yi (2 Qiao quantization factor e is the exponential function 2 dead according to different criteria, different values ​​of?;

A(2。用于调整预测模式在输出码流中所占的比例,e尸越大,预测模式所占比例越高;s(x,y)与s'(x,y)分别表示原始点和预测点,x、 y为其坐标值;根据相邻块对当前块预测,得到当前块的最可能预测模式,R用来区别是否为最可能预测模式,当前模式是最可能的预测模式时及=0,其他8种情况下i? = l; A (2 prediction mode for adjusting the proportion, the greater the corpse in the output stream e, the higher the proportion of the prediction mode;. S (x, y) and s' (x, y) represent the original point and prediction point, x, y coordinate values ​​for; the neighboring blocks of the current block is predicted, the most likely prediction mode of the current block, R is used to distinguish whether the most likely prediction mode, the current mode is the most likely prediction mode and = 0, i other eight situations = l?;

4、重复l-3步,从9种预测方式中选择。 4, repeat steps l-3, selected from the 9 prediction mode. 对16值最小的一个,即为最佳的预测模式。 A minimum value of 16, that is, the best prediction mode.

虽然这种全搜索方法可以找到最佳的预测模式,但是它的计算量非常大,是帧内和帧间编码中耗时很多的一部分。 Although this full search of ways to find the best prediction mode, but it is very computationally expensive, it is part of a lot of intra and inter coding time-consuming.

专利CN200410006340公开的方法,通过计算待编码宏块的紋理特征, 然后根据紋理特征中的紋理方向选出最优的预测模式。 The method disclosed in the patent CN200410006340, texture features by computing a macroblock to be encoded, and then the optimum prediction mode is selected according to the texture direction of the texture features. 这种方法需要进行宏块紋理分析,如灰度梯度法、傅立叶频谱分析法等,计算复杂度较高,不利于系统后期的平台优化。 This approach requires macroblock texture analysis, such as intensity gradient method, the Fourier spectrum analysis method or the like, high computational complexity, is not conducive to optimizing the system platform later.

专利CN200480006978^Hf的方法,通过计算帧内编码块的边缘方向信息,然后进行最优预测模式的选择。 Patent CN200480006978 ^ Hf method, by calculating the edge direction information of the intra-coded blocks, and select an optimal prediction mode. 这种方法需要确定块内所有象素的边缘矢量的幅值和角度,为每一个象素计算边缘方向直方图,计算量很大,实用性不高。 This approach requires determining the magnitude and angle of an edge vector for all pixels within the block, calculates an edge direction histogram for each pixel, a large amount of calculation, practicality is not high. 发明内容 SUMMARY

有鉴于此,本发明的主要目的在于提供一种实用性强,计算简便的低复杂度的帧内预测模式选择方法。 In view of this, the main object of the present invention is to provide a strong, practical availability, calculation is simple low complexity intra prediction mode selection method.

为达到上述目的,本发明的技术方案是这样实现的: 一种低复杂度的帧内预测模式选择方法,在当前块中取表征紋理特征的两组样本点,根据相邻块的开销值,计算阈值,该方法包括以下步骤: To achieve the above object, the technical solution of the present invention is implemented as follows: A low complexity intra prediction mode selection method, two samples taken characterizing texture feature points in the current block, based on cost values ​​of adjacent blocks, calculating the threshold value, the method comprising the steps of:

A、 判断当前块是否处于图像帧的特殊位置,如果是,计算该特殊位置下的最优预测模式并输出,结束计算;如果不是,进入步骤B; A, determines whether the current block is the special position of the image frame, and if so, computing the optimal prediction mode at this particular location, and outputs the calculation is finished; if not, goes to step B;

B、 计算当前块在最可能预测模式下的开销值,如果小于阈值,则输出该模式,结束计算;反之,进入步骤C; B, calculates the cost value of the current block in the most likely prediction mode, if less than the threshold, the output of the pattern, the end of the calculation; otherwise, goes to step C;

C、 选择所有非最可能的预测模式中开销值最小的一个,作为临时最优预测模式; C, most non-select all possible prediction modes in a minimum cost value, the optimum prediction mode as a temporary;

D、 比较最可能预测模式与临时最优预测模式的开销值,选择其中较小的作为最优预测模式输出。 D, and comparing the most likely prediction mode cost value of the optimal prediction mode temporary, select the smaller as the optimal prediction mode output.

步骤A中特殊位置是指,当前块处于图像帧的最左上方、最上方或最左方。 Step A special position is a position, in the top left block of the current image frame, top or left. 步骤A所述计算该特殊位置下的最优预测模式分三种情况: 当前块位于图像帧最左上方,最优预测模式为2; A step of calculating the optimal prediction mode at that particular position in three cases: a current block located top left image frame, the optimum prediction mode is 2;

当前块在图像帧的最上方,计算在模式1、模式2和模式8下,所述两组样本点的开销值,选择开销值最小的模式作为当前块的最优预测模式; At the top of the current block of the image frame, 1, 8 in the calculation mode, mode 2 and mode, the cost of the two sample points, selects the smallest cost value as the optimum mode is a prediction mode of a current block;

当前块在图像帧的最左方,计算在模式0、模式2、模式3和模式7下,所述两组样本点的开销值,选择开销值最小的模式作为当前块的最优预测模式。 In the far left of the current block of the image frame, calculate 0, mode 2, mode 3, and mode 7 modes, the cost of the two sample points, selects the smallest cost value as the optimum mode is the prediction mode of the current block.

步骤C中计算临时最优预测模式开销值的步骤包括: The step of calculating the cost value of the temporary optimum prediction mode in the step C comprises:

Cl、计算所有非最可能的预测模式下第一组样本点的绝对误差和,绝对误差和最小的模式,记为临时最优预测模式M!, M! Cl, calculate the absolute error of the first set of sample points of all the non-prediction mode and the most probable, and minimum absolute error model, referred to as the temporary optimal prediction mode M !, M! 左方的模式为M2, M! Left mode M2, M! 右方的模式为M3; Right mode M3;

C2、判断模式2是否为最可能的预测模式,如果是,记M4为空,进入步骤C4;否则,记模式2为MU,进入步骤C3; C2, determines whether the mode 2 the most likely prediction mode, and if so, denoted M4 is empty, proceeds to step C4; otherwise, referred to as a MU mode 2, the step of entering a C3;

C3、判断M2是否为最可能的预测模式,如果是,重新记M2左方的模式为M2,进入步骤C4;否则,判断M3是否为最可能的预测^^莫式,如果是,重新记M3右方的模式为M3,进入步骤C4; C3, M2 is determined whether the most likely prediction mode, and if so, the left credited M2 mode M2, proceeds to step C4; otherwise, it is determined whether M3 is most likely prediction ^^ Mohs, and if so, re-credited M3 right mode M3, proceeds to step C4;

C4、分别计算才莫式Mp M2、 M3和线的第二组样本点的绝对误差和,并与相应的第一组样本点得绝对误差和相加,得到各才莫式的绝对误差和,选#^色对误差和最小的模式作为临时最优预测模式,进入步骤D。 C4, respectively, was calculated second set of samples Mohs Mp point M2, M3 and line and absolute error, and to obtain the absolute error corresponding to the first set of sample points and addition, to obtain the respective absolute error only and a Mohs, # ^ is selected from the color mode and the minimum error as a provisional optimal prediction mode, proceeds to step D.

根据图像块的紋理特征,两组样本点包含的点数相同或不同。 The texture features of the image block, comprising two samples of the same point or different points.

所述方法适用于最小块为4 x 4的图像帧。 The method is suitable for the minimum block of 4 x 4 image frames.

本发明通过选取特征点代替整个数据块,在对视频编码质量没有明显影响的情况下,显著降低帧内预测模式选择的计算量,提高视频编码的实时性。 The present invention replace the entire block of data by selecting feature points, without significant effect on the quality of video coding, the significant reduction amount calculating an intra prediction mode selection, improve real-time video encoding.

附图说明 BRIEF DESCRIPTION

图1为4x4块预测样本点示意图; Figure 1 is a schematic view point of 4x4 block prediction samples;

图2为4x4块预测的8种预测方向示意图; FIG 2 is a schematic block 8 4x4 kinds of prediction directions of the prediction;

图3为相邻块A、 B、 C关系图; FIG 3 is a neighboring blocks A, B, C diagram;

图4是本发明所述方法的流程图。 FIG 4 is a flowchart of the method of the invention.

具体实施方式 Detailed ways

本发明的主要思想是:从简化开销值(Cost)的计算函数和减少需要验证的模式数两方面入手,对帧内预测模式选择进行处理。 The main idea of ​​the invention is: to start from both simplified cost value (Cost) calculation function and reduce the need for validation number of modes, intra prediction mode selected for processing. 通常一个4x4的亮度块具有的细节较少,紋理结构是比较平滑的,这样就可以通过减少4x4亮度块中需要计算的样本点数目来简化Cost值计算函数。 Usually less 4x4 luminance block has a details, the texture is relatively smooth, so that it can simplify Cost value calculating function by reducing the number of sample points in the 4x4 luma block needs to be calculated. 另外,在基于4x4 亮度块的帧内预测中,最优的预测模式通常与次优的预测模式具有相似的方向;并且当QP很大时(比如40),其他模式中的罚因子4;M:QP)将会增大, 所以最可能的预测模式通常就是最优的预测模式;特别的,在图像内容非常混杂的区域,任何一种预测模式都不可能达到很好的预测效果。 Further, the prediction based on the intra 4x4 luminance block, the optimal prediction mode having a generally similar direction to the sub-optimal prediction mode; and when the QP is large (such as 40), the other mode penalty factor 4; M : QP) will increase, so the most likely prediction mode is usually the best prediction mode; in particular, the image content in very mixed area, any kind of prediction model can not achieve good prediction. 以图l所示的4x4块为例,本方法的当前块(C块)预测的主要步骤如下: In the 4x4 block shown in FIG. L as an example, the main step of predicting the current block (C block) of the present method are as follows:

步骤1:在当前块中取表征紋理特征的两组样本点。 Step 1: Take two sample points characterizing the texture features in the current block.

取当前4x4块中的e、 f、 g、 h、 m、 n、 o 和p作为第一组样本点,a、 b、 c、 d、 i、 j、 k和l作为第二组样本点。 Take the current 4x4 block e, f, g, h, m, n, o and p are the first set of sample points, a, b, c, d, i, j, k and l as a second sample point. 在计算Cost值时就采用公式: Co《=&4Z)„ + ( 3 ) Cost value is when calculating using the formula: Co "= & 4Z)" + (3)

其中,SADn是利用每一组中的n个样本点计算得到的绝对误差和,R 和X(QP)与等式(1)中的含义相同。 Wherein, SADn n is the use of sample points in each group and calculated absolute error, R and X (QP) of the same meaning as in equation (1).

上述两组样本点Costm和Costn2的和,记为Costm; SADnl和SADn2的和,记为SADm。 The above two sets of sample points and Costm and Costn2, referred to as Costm; SADnl and SADn2 and, referred to as SADm.

步骤2:判断当前块是否处于图像帧的特殊位置,如果是,计算该位置下的最优预测模式并输出,退出计算;如果不是,进入步骤3。 Step 2: determining whether the current block is the special position of the image frame, and if so, computing the optimal prediction mode and outputs the position, calculated exit; if not, proceeds to step 3.

所谓图像帧的特殊位置是指当前块处于图像帧的最左上方、最上方或最左方。 Image frame position of a so-called special means in the top left of the current block of image frame, top or left. 其中,图像帧最左上方的预测模式为2;如果当前块在图像帧的最上方, 则在模式l、模式2和模式8下,计算上述两组样本点的Costm值,选择CosU 值最小的一个模式作为当前块的最优预测模式,并退出计算;如果当前块在图像帧的最左方,则在模式0、模式2、模式3和模式7下,计算上述两组样本点 Wherein the most upper left image frame prediction mode is 2; if the current block at the top of the image frame, in the L mode, mode 2 and mode 8, Costm calculated value of the two sample points, a minimum value selection CosU a mode as the optimal prediction mode of the current block is calculated and exit; if the current block in the far left image frame, then in mode 0, mode 2, mode 3 and mode 7, the calculation of the two sample points

的COSU值,选择C0Stm值最小的一个模式作为当前块的最优预测模式,并退 The COSU value, a minimum value selection C0Stm mode as the optimal prediction mode of the current block and back

出计算。 The calculation. 如果当前块不是处于图像帧的特殊位置,则ii7v步骤3。 If the current block is not a particular position in the image frame, the ii7v step 3. 步骤3:根据相邻块的Costm值,计算阈值T。 Step 3: The Costm values ​​of neighboring blocks, to calculate the threshold T.

相邻块位置如图3所示,每个块计算之后都会保存其C0Stm值,从中读 Adjacent to the block position shown in Figure 3, which will be saved after each block is calculated C0Stm value, read out therefrom

取A块、B块的Costm值。 A block taking, Costm value B block. 根据A块、B块的特征预测C块,得到C块的最可能预测模式,记录该模式。 The block A, the block B wherein C-block prediction, the most likely prediction mode of the block C, the recording mode. 如果最可能的预测模式就是最优预测模式, 则C块与A块或B7 = min(C《,CWmS ) + ,) ( 4 ) If the most likely prediction mode is the optimal prediction mode, the block A or the block C B7 = min (C ", CWmS) +,) (4)

其中,CostmA是A块的Costm值,CostmB是B块的Cos^值,X(QP)与等式(1 )中的含义相同。 Wherein, CostmA Costm value of A is block, CostmB Cos ^ value B is the block, the same as X (QP) in Equation (1) in the meaning.

步骤4:计算C块在最可能的预测模式下的CosW值,记为CostmC。 Step 4: Calculated C block CosW value in the most likely prediction mode, referred to as CostmC. 此时,R= 0,根据(1 )式,CostmC = SADm。 In this case, R = 0, according to (1), CostmC = SADm. 如果Costmc的结果小于阈值T, 那么就把它作为当前4x4亮度块的最优预测模式,退出计算,将最可能的预测模式作为最终的结果输出;否则说明最可能的预测模式不能像预测A块或B块的良好效果一样来预测C块,就必须在其他的8种预测模式中进行挑选,进入步骤5。 If the result is less than Costmc threshold T, then put it as the optimal prediction mode of the current 4x4 luminance block, exit calculation, the most likely prediction mode as the final result output; otherwise, this most likely prediction mode is not the prediction image block A or C block is predicted with good results as B-block, the other must be selected in eight kinds of prediction modes, proceeds to step 5.

步骤5:通过计算其他8种模式下第一组样本点的SADnl,选出临时最优预测模式及其左右模式。 Step 5: calculating a first set of sample points SADnl other 8 modes, the optimal prediction mode is selected and the temporary mode around.

除了最可能的预测模式外,其他预测模式在计算Costn值时都要受到罚因子X(QP)的相同制约,所以在比较其他8种预测才莫式的预测效果时,可以先不考虑罚因子的影响。 But the most likely prediction mode, a prediction mode when calculating the other value will be subject to the same constraints Costn penalty factor X (QP), so that when comparing only eight other prediction formula Mo prediction effect, may not considered penalty factor Impact. 由于模式2与方向性无关,需要单独考虑,将计算得到的模式与模式2、最可能预测模式综合比较,得到临时最优预测模式及其左右模式。 Because nothing to do with the directional pattern 2, need to be considered separately, the calculated mode and Mode 2, the most likely prediction mode comprehensive comparison, obtain optimal prediction mode and about the temporary mode. 具体步骤如下: Specific steps are as follows:

步骤501:计算所有非最可能的预测模式下第一组样本点的SADnl,得到SAD^值最小的一个模式,记为临时最优预测才莫式M,。 Step 501: calculating a first set of sample points SADnl all non-prediction mode most likely to yield a minimum value SAD ^ model, referred to as the optimal prediction only temporary Morse M ,. ]V^左方的模式记为M2, N^右方的模式记为M3。 ] V ^ referred to as the left mode M2, N ^ referred to as the right mode M3. 各种模式的左右关系如表1所示。 Relationship between about various modes as shown in Table 1.

<table>table see original document page 9</column></row> <table>表1 <Table> table see original document page 9 </ column> </ row> <table> TABLE 1

步骤502:因为模式2与方向性无关,表l中无法体现,所以在计算中单独考虑,用M4表示模式2与最可能预测模式的关系,判断模式2是否为最可能的预测模式,如果是,记M4为空,进入步骤6;否则,记模式2为M4, ii7v 步骤503。 Step 502: Because the directionality regardless of mode 2, can not be reflected in Table l, it is considered separately in the calculation, the relational model represents the best possible prediction mode, determines whether the mode 2 the most likely prediction mode M4, if , denoted M4 is empty, proceeds to step 6; otherwise, referred to as mode 2 M4, ii7v step 503.

步骤503:判断M2是否为最可能的预测模式,如果是,重新记M2左方的模式为M2,进入步骤6;否则,判断M3是否为最可能的预测模式,如果是,重新记M3右方的才莫式为M3,进入步骤6。 Step 503: Determine whether M2 is the most likely prediction mode, and if so, to re-record mode M2 ​​M2 left, proceed to step 6; otherwise, to determine whether M3 is the most likely prediction mode, and if so, the right to re-credited M3 the only Mohs as M3, proceed to step 6.

步骤6:分别计算在步骤5中得到的预测模式Mp M2、 M3和M4的第二组样本点的SADn2值,并与其各自的第一组样本点得SADnl值相加,选择SADm 值最小的一个作为新的临时最优预测模式,计算得到它的Costm值,记为Cost謹,进入步骤7。 Step 6: calculate the prediction mode Mp M2 obtained in step 5, M3 and SADn2 value M4 of the second set of sample points and their respective sample points obtained SADnl first set value is added, a minimum value selection SADm as a new temporary optimal prediction mode, Costm its calculated value, denoted Cost wish, proceeds to step 7.

步骤7:比较C0StmC和C0StmM,选择其中较小的一个作为最优预测模式输 Step 7: Compare and C0StmC C0StmM, select the smaller one as the optimal prediction mode input

出。 Out. 如果二者相等,说明图像的方向细节已经发生了变化,此时临时最优预测模式的预测效果要优于最可能的预测模式。 If they are equal, indicating the direction of image detail has changed, then the temporary optimal prediction mode prediction effect is superior to the best possible prediction mode. 这时选择临时最优预测模式作为最终的最优预测^^莫式,那么对于后续右方和下方的块具有更佳的预测效果。 In this case the temporary optimal prediction mode selected as the final optimum prediction ^^ Mohs, then a predicted block for a better effect subsequent right and below.

在全搜索方法下为了找到最优预测模式,每个4x4块需要计算16x9=144 个样本点的Cost值。 In the full search method to find the optimal prediction mode for each 4x4 block needs to be calculated value Cost 16x9 = 144 sample points. 而实施例中的快速搜索方法最好的情况是只需要计算步骤1~4,即最可能的预测模式下16个点的Cost值。 And the best method in the case where a quick search is only necessary to calculate embodiment Cost value of the 16 points at step 1-4, i.e. the most likely prediction mode. 经实验发现,当QP值增大时,只计算最可能预测模式的概率也随之增大。 The experiment found that when the QP value increases, only to calculate the probability of the most likely prediction mode increases. 例如在QP=16时约为30%, QP=31时增大为50 % , QP=48时最大接近80 % 。 For example about 30% QP = 16, QP = 50% increased to 31, QP = 48 to 80% of maximum approach.

最差的情况是步骤5中需要计算模式2, Cost值的样本点总数为: (16* 1) +(8* 8)+ (8* 4) =112 Worst case calculation mode in step 2 needs to 5, the total number of sample points Cost value is: (16 * 1) + (8 * 8) + (8 * 4) = 112

可以看出即使在最差的情况下,实施例中的快速模式搜索方法还是节省了32个样本点的计算量,使得预测模式选择的时间效率提高20%左右。 It can be seen even in the worst case, the implementation of the fast search mode embodiment is a method saves computation 32 sample points, so that the efficiency of the selected prediction mode time increased by 20%.

步骤1中,两组样本点包含的点数可以不同,根据图像块的紋理特征, 选取最能表征紋理特征的点,作为样本点。 In Step 1, the two points comprises sample points may be different, according to the texture features of the image block, selecting feature points best represents the texture as a sample point. 步骤3中阈值T的选取至关重要:如果T较小,则不能有效的减少需要计算的预测模式数;否则,较容易选择最可能的预测模式作为最终的结果,不能有效的找到当前块的最佳预测模 Step 3 threshold of the critical T: If T is small, can not effectively reduce the number of prediction modes to be calculated; otherwise, it is easier to select the most likely prediction mode as the final result, can not effectively find the current block best prediction mode

式。 formula. 一般的,在图像内容比较复杂的区域,可以适当的增大T的值;否则, 应该减小T的值。 In general, the more complex the image content region may be appropriate to increase the value of T; otherwise, the value of T should be reduced. 另外,T的选择与QP值的大小也有关系:当QP较大时, 预测误差较大,则应该适当的增大T的值;否则,应该减小T的值。 Further, the size and selection of the QP value is also related to T: When QP is larger, the prediction error is large, increasing the value of T should be appropriate; otherwise, the value of T should be reduced.

以上所述,仅为本发明的较佳实施例而已,并非用于限定本发明的保护范围。 The above are only preferred embodiments of the present invention but are not intended to limit the scope of the present invention.

Claims (5)

1、一种低复杂度的帧内预测模式选择方法,其特征在于,在当前块中取表征纹理特征的两组样本点,根据相邻块的开销值,计算阈值,该方法包括以下步骤: A、判断当前块是否处于图像帧的特殊位置,如果是,计算该特殊位置下的最优预测模式并输出,结束计算;如果不是,进入步骤B; B、计算当前块在最可能预测模式下的开销值,如果小于阈值,则输出该模式,结束计算;反之,进入步骤C; C、选择所有非最可能的预测模式中开销值最小的一个,作为临时最优预测模式; D、比较最可能预测模式与临时最优预测模式的开销值,选择其中较小的作为最优预测模式输出; 所述步骤C中计算临时最优预测模式开销值的步骤包括:C1、计算所有非最可能的预测模式下第一组样本点的绝对误差和,绝对误差和最小的模式,记为临时最优预测模式M1,M1左方的 1, an intra-forecast mode selecting method of low complexity, characterized in that the two samples taken characterizing texture feature points in the current block, the cost value of adjacent blocks, calculate the threshold value, the method comprising the steps of: a, determines whether the current block is the special position of the image frame, and if so, computing the optimal prediction mode at this particular location, and outputs the calculation is finished; if not, goes to step B; B, calculates a current block in the most likely prediction mode the cost value is less than the threshold value, the output of the model, the calculation is finished; otherwise, goes to step C; C, select all non most probable prediction modes smallest cost value as the temporary optimal prediction mode; D, Comparative most provisional prediction mode may be optimal prediction mode cost value, selecting the smaller of the output as the optimal prediction mode; step C the calculated cost of the temporary optimal prediction mode comprises the step of: C1, calculating the most likely of all non- and absolute error, the minimum absolute error model and a first set of sample points in the prediction mode, referred to as the temporary optimal prediction mode M1, M1 left 模式为M2,M1右方的模式为M3;C2、判断模式2是否为最可能的预测模式,如果是,记M4为空,进入步骤C4;否则,记模式2为M4,进入步骤C3;C3、判断M2是否为最可能的预测模式,如果是,重新记M2左方的模式为M2,进入步骤C4;否则,判断M3是否为最可能的预测模式,如果是,重新记M3右方的模式为M3,进入步骤C4;C4、分别计算模式M1,M2、M3和M4的第二组样本点的绝对误差和,并与相应的第一组样本点的绝对误差和相加,得到各模式的绝对误差和,选择绝对误差和最小的模式作为临时最优预测模式,进入步骤D。 Mode M2, M1 right mode M3; C2, 2 determines whether the mode is the most likely prediction mode, and if so, denoted M4 is empty, proceeds to step C4; otherwise, referred to as mode 2 M4 proceeds to step C3; C3 , M2 is determined whether the most likely prediction mode, and if so, the left credited M2 mode M2, proceeds to step C4; otherwise, it is determined whether M3 is most likely prediction mode, and if so, re-credited right mode M3 M3 is, proceeds to step C4; C4, were calculated mode M1, the absolute error M2, M3 and M4 of the second set of sample points and, with the sum of absolute difference corresponding to a first set of sample points, obtained in each mode and absolute error, and selecting minimum absolute error mode as the temporary optimal prediction mode, proceeds to step D.
2、 根据权利要求1所述的低复杂度的帧内预测模式选择方法,其特征在于, 步骤A中特殊位置是指,当前块处于图像帧的最左上方、最上方或最左方。 2. The low complexity intra prediction mode of a selection method according to claim, wherein, in step A refers to a particular location, in the top left of the current block of the image frame, top or left.
3、 根据权利要求2所述的低复杂度的帧内预测模式选择方法,其特征在于, 步骤A所述计算该特殊位置下的最优预测模式分三种情况:当前块位于图像帧最左上方,最优预测模式为2;当前块在图像帧的最上方,计算在模式l、模式2和模式8下,所述两组样本点的开销值,选择开销值最小的模式作为当前块的最优预测模式;当前块在图像帧的最左方,计算在模式0、模式2、模式3和模式7下,所述两组样本点的开销值,选择开销值最小的模式作为当前块的最优预测模式。 3, a low complexity intra prediction mode of the selection method according to claim 2, wherein the step of calculating the optimal prediction mode A at this particular location in three cases: a current block is located in the uppermost left image frame side, the optimum prediction mode is 2; current block top image frame, calculate the L mode, mode 2 and 8, the cost of the two sample points, select the smallest cost value as the mode of the current block the optimal prediction mode; current block in the far left of the image frame, calculate in mode 0, mode 2, mode 3, and mode 7, the cost of the two sample points, select the smallest cost value as the mode of the current block optimal prediction mode.
4、 根据权利要求1所述的低复杂度的帧内预测模式选择方法,其特征在于, 根据图像块的紋理特征,两组样本点包含的点数相同或不同。 4. The low complexity intra prediction mode of a selection method according to claim, characterized in that, in accordance with the texture features of the image block, two samples contained the same point or different points.
5、 根据权利要求1所述的低复杂度的帧内预测模式选择方法,其特征在于, 所述方法适用于最小块为4x4的图像帧。 5, the intra prediction mode according to a low complexity of the selection method of claim 1, characterized in that the method is applicable to most of the 4x4 small blocks of the image frame.
CN 200710175906 2007-10-15 2007-10-15 Low-complexity in-frame estimation mode selection method CN100551075C (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN 200710175906 CN100551075C (en) 2007-10-15 2007-10-15 Low-complexity in-frame estimation mode selection method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN 200710175906 CN100551075C (en) 2007-10-15 2007-10-15 Low-complexity in-frame estimation mode selection method

Publications (2)

Publication Number Publication Date
CN101175212A CN101175212A (en) 2008-05-07
CN100551075C true CN100551075C (en) 2009-10-14

Family

ID=39423419

Family Applications (1)

Application Number Title Priority Date Filing Date
CN 200710175906 CN100551075C (en) 2007-10-15 2007-10-15 Low-complexity in-frame estimation mode selection method

Country Status (1)

Country Link
CN (1) CN100551075C (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101835048A (en) * 2010-03-30 2010-09-15 香港应用科技研究院有限公司 Method and device for performing video coding by a just noticeable difference model based on ABT

Families Citing this family (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101668202A (en) * 2008-09-01 2010-03-10 中兴通讯股份有限公司 Method and device for selecting intra-frame prediction mode
CN101895739B (en) 2009-05-20 2012-12-19 深圳市融创天下科技股份有限公司 Block statistical characteristic-based block encoding method
CN102196255B (en) * 2010-03-11 2013-04-10 中国科学院微电子研究所 Formation method of video encoding complexity control model
CN102300088B (en) * 2010-06-25 2013-11-06 财团法人工业技术研究院 In-frame prediction mode optimization method as well as image compression method and device
CN102186081B (en) * 2011-05-11 2013-09-18 北京航空航天大学 H.264 intra-frame mode selection method based on gradient vector
CN105245894B (en) * 2011-06-17 2017-08-08 Jvc建伍株式会社 Picture decoding apparatus, picture decoding method and reception device, method of reseptance
PL3136728T3 (en) 2011-06-28 2018-05-30 Samsung Electronics Co., Ltd. Method for decoding video with intra prediction
CN102857750B (en) * 2011-07-01 2015-05-06 华为技术有限公司 Method, device and system for in-frame predictive coding and encoding
KR20130049523A (en) * 2011-11-04 2013-05-14 오수미 Apparatus for generating intra prediction block
CN103731670B (en) * 2013-12-25 2017-02-01 同观科技(深圳)有限公司 Intra-frame prediction algorithm of image
CN104954787B (en) * 2014-03-27 2018-03-23 浙江大华技术股份有限公司 HEVC inter-frame forecast mode selection methods and device
CN106162175A (en) * 2015-03-26 2016-11-23 北京君正集成电路股份有限公司 Method for choosing frame inner forecast mode and device
CN106961606A (en) * 2017-01-26 2017-07-18 浙江工业大学 The HEVC intra-frame encoding mode systems of selection of feature are divided based on texture

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1535027A (en) 2004-01-16 2004-10-06 北京工业大学 Inframe prediction method used for video frequency coding
CN1585495A (en) 2004-06-11 2005-02-23 上海大学 Quick selection of prediction modes in H.264/AVC frame
CN1753500A (en) 2005-10-31 2006-03-29 连展科技(天津)有限公司 Method of selecting in frame prediction mode based on H.264/AVC standard frame image
CN1756364A (en) 2004-09-30 2006-04-05 华为技术有限公司 Method for selecting intra-prediction mode

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1535027A (en) 2004-01-16 2004-10-06 北京工业大学 Inframe prediction method used for video frequency coding
CN1585495A (en) 2004-06-11 2005-02-23 上海大学 Quick selection of prediction modes in H.264/AVC frame
CN1756364A (en) 2004-09-30 2006-04-05 华为技术有限公司 Method for selecting intra-prediction mode
CN1753500A (en) 2005-10-31 2006-03-29 连展科技(天津)有限公司 Method of selecting in frame prediction mode based on H.264/AVC standard frame image

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101835048A (en) * 2010-03-30 2010-09-15 香港应用科技研究院有限公司 Method and device for performing video coding by a just noticeable difference model based on ABT

Also Published As

Publication number Publication date
CN101175212A (en) 2008-05-07

Similar Documents

Publication Publication Date Title
CN1220391C (en) Image encoder, image decoder, image encoding method, and image decoding method
CN104780364B (en) Determine the intra prediction mode of image coding unit and image decoding unit
CN1934871B (en) Method of mode decision for inter prediction in video coding and system
CN102484704B (en) Method and apparatus for encoding video, and method and apparatus for decoding video
CN101023676B (en) Method and apparatus for motion vector prediction in temporal video compression
CN100566411C (en) Method, medium, and filter removing a blocking effect
CN100401789C (en) Quick selection of prediction modes in H.264/AVC frame
CN101945279B (en) Method and device for selecting predictive modes for pixel group for encoding video image
CN101267562B (en) Video processing system and device with encoding and decoding modes and method for use therewith
CN101502119B (en) Adaptive geometric partitioning for video decoding
CN101494782B (en) Video encoding method and apparatus, and video decoding method and apparatus
CN101686393B (en) Fast-motion searching method and fast-motion searching device applied to template matching
CN101379816B (en) Temporal error concealment for bi-directionally predicted frames
RU2607233C2 (en) Method and apparatus for encoding and decoding image using intra-frame prediction
CN101352046B (en) Image encoding/decoding method and apparatus
CN101513071B (en) Method and apparatus for determining expected distortion in decoded video blocks
CN101529916B (en) Video encoding with intra encoding selection
CN101507277B (en) Image encoding/decoding method and apparatus
CN1929611A (en) Methods and apparatus for video intraprediction encoding and decoding
CN101978697B (en) Coding mode selection for block-based encoding
CN102172022A (en) Digital video coding with interpolation filters and offsets
CN100461867C (en) Inage predicting encoding method in frame
CN1275469C (en) Method for pridicting sortable complex in frame
CN102792697A (en) Method and apparatus for encoding and decoding motion vector
CN101860754B (en) Method and device for coding and decoding motion vector

Legal Events

Date Code Title Description
C06 Publication
C10 Entry into substantive examination
C14 Grant of patent or utility model
CF01