CN100401789C - Quick selection of prediction modes in H.264/AVC frame - Google Patents

Quick selection of prediction modes in H.264/AVC frame Download PDF

Info

Publication number
CN100401789C
CN100401789C CN 200410025123 CN200410025123A CN100401789C CN 100401789 C CN100401789 C CN 100401789C CN 200410025123 CN200410025123 CN 200410025123 CN 200410025123 A CN200410025123 A CN 200410025123A CN 100401789 C CN100401789 C CN 100401789C
Authority
CN
China
Prior art keywords
mode
block
luminance
decision
i4
Prior art date
Application number
CN 200410025123
Other languages
Chinese (zh)
Other versions
CN1585495A (en
Inventor
颖 张
张兆扬
滕国伟
Original Assignee
上海大学
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 上海大学 filed Critical 上海大学
Priority to CN 200410025123 priority Critical patent/CN100401789C/en
Publication of CN1585495A publication Critical patent/CN1585495A/en
Application granted granted Critical
Publication of CN100401789C publication Critical patent/CN100401789C/en

Links

Abstract

本发明涉及一种H.264/AVC实时编码器的帧内预测模式快速选择方法。 The present invention relates to an H.264 / AVC encoder in real time Fast intra-prediction mode selection method. 在H.264/AVC编码器中,帧内编码采用帧内预测方法,利用色度块和亮度块、亮度块的内部以及亮度块预测模式选择间的相关性,根据率失真优化(RDO)准则确定不同判决依据(RDO或SAD)来快速选择帧内预测模式。 In H.264 / AVC encoder, intra-coded intra-frame prediction method, using the luminance blocks and chrominance blocks, the correlation between the internal blocks of luminance blocks and luminance prediction mode selection, optimization (RDO) according to a rate-distortion criterion to determine the different decision based on (RDO or SAD) to quickly select an intra-prediction mode. 本发明提供的方法,对比于全搜索方法,可以在信噪比和码率变化很小的同时,可使编码速度平均提高6倍以上。 The method of the present invention provides, in contrast to a full search method, may be in a small variation rate and SNR same time, can increase the speed to more than 6 times.

Description

H. 264/AVC帧内预测模式的快速选择方法技术领域本发明涉及一种信息处理的视频编码的方法,特别是一种H. 264/AVC帧内预测模式的快速选择方法。 TECHNICAL FIELD The quick selection H. 264 / AVC intra prediction mode of the invention relates to a video encoding an information processing method, in particular fast method for selecting a H. 264 / AVC intra prediction mode. 背景技术自从ITU-T颁布了H. 261后,基于块的混合编码方法作为框架被普遍采用并不断发展。 BACKGROUND ART Since issued ITU-T H. 261, block-based hybrid coding method is widely used as a framework and continue to develop. 国际标准H.261、 H.263、 MPEG-1、 MPEG-2、 MPEG-4等的制定极大推动了多媒体技术的应用。 Development of international standards H.261, H.263, MPEG-1, MPEG-2, MPEG-4, etc. greatly promoted the application of multimedia technology. 然而,随着新服务的大量增加和高清电视的即将普及,以及现有传输媒体如Cable Modem、 Xdsl、 UMTS较低的传输码率,迫切需要提高编码效率;同时还要求视频编码标准能够适应现存或未来的网络。 However, with the substantial increase in the popularity of the upcoming new services and high-definition television, as well as existing transmission media such as Cable Modem, Xdsl, UMTS lower transmission rate, the urgent need to improve the coding efficiency; also asked to adapt existing video coding standard or future network. 这些问题为新视频编码标准H. 264/AVC的产生提供了切实的依据。 These issues provide a practical basis for the new video coding standard H. produce 264 / AVC's. 这些视频编码标准采用混合编码框架,即将输入视频图象分割成宏块,对其进行预测、变换、量化、熵编码。 These video coding standards using hybrid coding framework, i.e. an input video image is divided into macroblocks, its prediction, transformation, quantization, entropy encoding. 预测包括帧内预测和帧间预测,帧内预测可去除空间冗余,而帧间预测可去除时间上的冗余。 Prediction including intra prediction and inter prediction, intra prediction may remove spatial redundancy, and interframe prediction time redundancy can be removed. H. 264/AVC仍遵从这种基于块的混合编码方案,但包含了许多新特征:VCL层和NAL层分离、空间域的帧内估计、1/4像素精度运动估计、自适应块尺寸运动补偿、 多参考帧运动补偿预测、通用B帧的概念、低复杂16bit的整数变换和量化、环内去块滤波器和高效的熵编码。 H. 264 / AVC still comply with such a hybrid block-based coding scheme, but includes many new features: VCL NAL layer and the layers were separated, the spatial domain of an intra estimation, 1/4-pixel accuracy motion estimation, motion adaptive block size compensation, multiple reference frame motion compensation prediction, the concept of generic B frames, a low complexity 16bit integer transform and quantization, and in-loop deblocking filter efficiency of entropy coding. 这些新特征一方面能够显著提高编码效率;另一方面具有网络友好性,可有效用于各种网络和应用环境。 These new features on the one hand the coding efficiency can be significantly improved; on the other hand has a network friendly, effective for a variety of network environments and applications. 其中H. 264/AVC为提高帧内图象的编码效率,采用帧内预测去除空间冗余性。 Wherein the H. 264 / AVC to improve the coding efficiency of an intra picture, the intra prediction using the spatial redundancy is removed. 这种帧内预测技术依然以宏块为处理单元,即通常将一帧图象分割成16x16的图象块进行编码。 This technology is still the intra prediction macroblock processing units, i.e., typically a 16x16 image into image blocks is coded. 然而基本处理块单元相对于其它标准由8x8降为4x4,即变换采用4x4的变换矩阵。 However, the basic process unit with respect to other standard block of 8x8 4x4 reduced, i.e. 4x4 transform using the transformation matrix. 因此亮度宏块支持两种帧内预测块尺寸:4x4和16x16,而色度仅支持8x8。 Thus luminance macroblock supports two intra prediction block size: 4x4 and 16x16, chrominance only supports 8x8. 每个进行帧内预测的当前块利用邻块的相邻像素值来预测当前块像素值,然后计算残差、变换、熵编码。 For each current block using the intra prediction of pixel values ​​of neighbor blocks adjacent to the current block the prediction pixel value, and then calculates a residual transform, entropy encoding. H. 264/AVC标准中每个4x4块的帧内预测包括9种预测模式(如附图l)。 H. 264 / AVC standard, intra prediction for each 4x4 blocks comprising 9 prediction modes (e.g., figures l). 而亮度16xl6和色度8x8仅支持4种预测模式,分别为:垂直(V)、水平(H)、直流(D)和平坦模式(P),分别简称为V、 H、 D和P模式。 And the luminance and chroma 8x8 16xl6 only supports 4 prediction modes, namely: a vertical (V), horizontal (H), direct (D) and the flat pattern (P), abbreviated as V, H, D, and P mode. 同时这项技术也被应用到帧间编码之中,即对当前宏块进行完帧间预测后,接着对它进行帧内预测,以率失真优化为最终判决依据,率失真小的模式被确定为最终模式。 Parallel to this technique has also been applied to inter-coded into, i.e., the current macroblock is finished after the inter prediction, intra prediction and then it, in order to optimize the rate-distortion is based on a final decision, the rate distortion model is determined for the final model. 这样既可以提高帧间编码宏块的编码效率,又可以提高容错能力。 This is to improve the encoding efficiency of inter-coded macroblocks, but also can improve fault tolerance. 例如对某帧(I帧、P帧或B帧皆可)中当前宏块M进行帧内预测编码,利用相邻块与当前块的紧挨的像素值来预测当前块的像素值,然后用当前块的像素值减去预测值得到残差矩阵,将此残差矩阵经过变换、量化、熵编码得到该宏块的编码码流。 For example, in a frame (I frame, P frame or B frame can be) the current macroblock M intra prediction encoding by using the pixel values ​​of the current block of a current block immediately to predict pixel values ​​of adjacent blocks, followed by subtracting the pixel values ​​of the current block to the prediction residual matrix worth, this residue matrix after the transformation, quantization, entropy encoding to obtain the macroblock encoded stream. 但利用邻块像素值的预测方式是不同的,因此产生了多种预测模式(如附图l)。 However, the prediction mode using the pixel values ​​of neighbor blocks are different, thus creating a plurality of prediction modes (e.g., figures l). 在判断这些模式中的最优模式,需要一个衡量的准则,目前采用了率失真优化(RD0-rate distortion optimization),其定义如下:〜謹(&,A 12,4。。£) = 0虹(^,/* (0 + ^。促^a:O^,A i0 公式(1)式中"ffic表示编码后总的失真,i?处c表示总的码率。;^。泥》o为拉格朗曰参数,通过确定编码选择项/,,使编码后的失真和码率总和最小。本发明在色度宏块和亮度16x16块的模式判决中采用了一种简化的判决准则为SAD (Sum of Absolute Difference),其定义如下:= —/'(U')) 公式(2)i=0产o式中/"刀为原始信号,/'(!•,/)是帧内预测后的预测信号,N等于8(色度块) 或16 (亮度块)。H. 264/AVC校验模型对于当前宏块帧内预测模式的确定给出了一种全搜索方法, 以色度宏块的帧内预测模式为外循环,以亮度宏块的两种块模式为内循环,即对应每一种色度宏块的帧内预测模式,都要分别比较亮度宏 In these modes is determined optimum mode, a criterion to measure the current using rate distortion optimization (RD0-rate distortion optimization), which is defined as follows: ~ Move (&, A 12,4 .. £) = 0 Rainbow . (^, / * (0 + ^ pro ^ a: O ^, a i0 formula (1) wherein "ffic represents the total distortion after encoding, I c represents the total rate; ^ mud?.." O It is said Lagrangian parameter, by determining the encoding selection / ,, the distortion and bit rate after encoding the smallest sum. the present invention employs a simplified decision criterion in the luminance and chrominance macroblock mode decision block is 16x16 SAD (Sum of Absolute Difference), which is defined as follows: = - / '(U')) equation (2) i = 0 where production o / "knife original signal, / '(•, /) is an intra! predicted prediction signal, N is equal to 8 (chrominance block) or 16 (luminance block) .H. 264 / AVC calibration model for determining the intra prediction mode of the current macroblock given a full search method, Israel macroblock of an intra prediction mode of the outer loop, in both the block mode of the luminance macroblock cycle, i.e., corresponding to each macroblock of an intra chrominance prediction mode, respectively, must compare the luminance macro 14和116的所有模式。故这种方法能够扫描到所有的模式,因此编码质量最高,运算量也是最大的。搜索的模式总数为4 x ( 16 x 9 + 4 ),共592个,这意味着为了得到一个宏块的最佳模式,需要计算592个不同的J,、,。w 。这使H. 264/AVC在现有的硬件条件下实现编码是不可能的。 发明内容本发明的目的是提供一种H. 264/AVC实时编码器的帧内预测椟式的快速选择方法,对比于全搜索方法,它可以在信噪比和码率变化很小的同时,可使编码速度平均提高6倍以上。为达到上述目的,本发明采用下述技术方案:一种H. 264/AVC实时编码器帧内预测模式的快速选择方法,在H. 264/AVC编码器中,帧内编码采用帧内预测方法,其特征在于利用当前宏块的色度块和亮度块帧内预测模式间的相关性、4x4亮度块14和16x16亮度块116的模式选择之间的相关性以及亮度块预测模式内部的相关性 All modes 14 and 116. Therefore, this method can scan all the modes, and therefore the highest encoding quality, is the largest amount of computation. Total search mode is 4 x (16 x 9 + 4), a total of 592, which means in order to get the best mode of a macroblock, 592 need to calculate different J ,,,. w. this allows the H. 264 / AVC encoding is not possible to achieve in the existing hardware conditions. SUMMARY oF tHE iNVENTION the present invention object to provide a H. 264 / AVC intra prediction of the real time encoder formula casket quick selection method, in contrast to a full search method, it can be a very small change in SNR and bit rate while the average encoding speed can Increases of more than 6 times to achieve the above object, the present invention adopts the following technical solution: a method for rapid selection H. 264 / AVC real time encoder intra prediction mode in H. 264 / AVC encoder, intra-coded intra-frame prediction method, characterized by using the correlation between luminance and chroma blocks of the current block is the intra prediction mode macroblock, and the correlation between the prediction mode of the luminance block luminance 4x4 luminance blocks 14 and block 116 of the selected 16x16 relevance of internal models ,以简化的率失真优化RDO为判决依据,快速进行帧内预测模式选择。快速选择的步骤如下:a) 确定色度块的模式选择具有独立性,即亮度宏块无论选择I16还是I4, 色度宏块模式都是一致的,从而首先对色度预测模式以SAD为准则进行独立判决,确定色度的最佳预测模式。 , A simplified rate-distortion optimization RDO decision is based, fast intra prediction mode selecting step to quickly select the following:. A) determining a block mode of the chrominance selection of independent, i.e., whether selection I16 luminance macroblock or I4, color of macroblock mode are the same, so that firstly chroma prediction mode for the SAD criterion independent decision to determine the best prediction mode chromaticity. b) 确定了色度的最佳预测模式后,为亮度快模式判决提供如下依据:对于亮度I16而言,当色度块选择直流D模式,亮度可以选择直流D、水平H、垂直V、平坦P模式;当色度块选择水平H模式,亮度可以选择直流D、水平H、平坦P模式;当色度块选择垂直V模式,亮度可以选择直流D、垂直V、平坦P模式;当色度块选择平坦P模式,亮度可以选择直流D、平坦P模式;而对于亮度块I4,亮度块I8选择的模式是亮度块14下绝大多数4x4块可能选择的模式;c) 对于宏块,14的选择概率远远大于I16,因此我们首先进行14下4x4块的模式判决;根据邻块的模式判断得到当前块的可能模式,并和由步骤b确定的可能模式,分别计算这两种模式下的RD0,并与阈值T0比较,小于T0,则可确定为当前4x4的最佳模式,否则较小值RD0的为可能最佳模式。 B) After determining the optimum prediction mode chroma, brightness mode offers faster decision based on the following: the luminance I16, when D chroma DC block selection mode, DC D may be selected brightness, horizontal H, vertical V, flat P mode; H level when the chroma block selection mode, DC D may be selected brightness, H level, planar model P; when the vertical V chrominance block mode selected, the luminance DC D may be selected, V the vertical, planar model P; when chromaticity P mode selection block flat, DC D may be selected brightness, the P mode is flat; for a luminance block I4, I8 luminance block selection mode is the mode 14 the vast majority of 4x4 blocks may be selected luminance blocks; c) for the macroblock 14 I16 is far greater than the probability of selection, we first performed at decision block 14 4x4 mode; determining a current block pattern may be obtained depending on the mode neighbor block and determined from step b, and possible mode, both modes are calculated the RD0, and compare with the threshold value T0, less than T0, it can determine the best mode for the current 4x4, otherwise the smaller value RD0 is probably the best mode. d) 比较步骤c中得到的可能最佳模式的预测方向邻近的模式的RD0,进一步确定可能的最佳模式;如果它是垂直V、水平H或直流D,则比较这三种模式中未曾比较的模式;如果它为其它模式,则比较所有其它未曾比较的模式;e) 在完成当前宏块的16个4x4块的模式判决后,计算宏块纹理复杂度,以MSE 来衡量,并与阈值T1比较,如果大于T1,提前终止判决,上述得到的结果为宏块最佳模式;如果小于T1,则进行亮度16x16块的模式判决;f) 根据步骤b中色度块的最佳模式与亮度块的关系,确定当前宏块116的可能模式;统计步骤e中14情况下16个子块的模式,得到概率最大的两种模式; 综合考虑这些可能模式,最终确定I16的可能模式。 RD0 adjacent modes best possible prediction direction mode d) obtained in the comparison step c, may further determine the best mode; if it is a vertical V, horizontal H or DC D, the comparison of the three modes never Comparison mode; if it is the other mode, comparing all modes other never comparison; E) after completion of the current macroblock 16 4x4 block mode decision, calculated macroblock texture complexity to MSE is measured, and a threshold T1 comparison, if greater than T1, early termination decision, the result of the best mode of the macro block; is less than T1, the 16x16 luminance block mode decision; F) according to the best mode in step b the luminance chrominance blocks relation block, a possible mode of the current macroblock 116; in the case of the statistical model in step e 14 16 sub-blocks, the probability of getting the maximum two modes; considering these possible modes, I16 finalized possible modes. 并以SAD为准则,确定116的最佳预测模式; g)自适应确定步骤e所述的阈值Tl来判断是否提前终止116判决和自适应确定步骤c所述的TO来判断是否提前终止14的判决。 SAD and criteria to determine the best prediction mode 116; G) determining an adaptive threshold Tl, step e to determine whether the early termination decision 116 determines adaptively TO and said step c to determine whether early termination of the 14 judgment. 所述的阈值通过自适应方法获取的具体步骤为: 阈值TO通过如下公式自适应获取:TO = "x ((2g-to -砂—麵W)/《謹/一coe/〖i"柳]则0])2其中,a是校正因子,其取值范围在(O,l], 卯—rnn = ^P%6 ,— 6te -砂一+ 15,砂—=(1«《—6to) / 6 ,《w"W — coe/是量化矢巨阵,QP为量化参数; 计算阈值n的自适应方法为: 初始化T1-K〈20;第一个宏块:依次进行14和116模式判决,得到二者的MSE (Minimum of Square Error),将均值赋给T1,作为下一宏块的阈值。 其它宏块:如果仅判决了14模式,则T1不变。如果依次进行了14和116模式判决, 则二者的MSE的均值赋给T1。本发明综合利用上述方法进行帧内预测模式选择。实验结果显示,可以提高编码速度数倍,且能够保证编码的码率和PSNR与全搜索方法相比变化不大。 附图说明图1是本发明的一个实施例的信息处理视频编码框图。 图2是本发明的一 The specific steps taken by an adaptive threshold method: TO adaptive threshold value obtained by the following formula: TO = "x ((2g-to - Sand - surface W) /" honor / a COE / 〖i "Liu] is 0]) 2 where, a is a correction factor, which is in the range (O, l], d -rnn = ^ P% 6, - 6te - a + 15 sand, sand - = (1 «" - 6to) / 6, "w" W - coe / downline vector quantization, QP is a quantization parameter; adaptive method for calculating the threshold value n: initialization T1-K <20; the first macro block: 14 and 116 sequentially mode decision to give the two MSE (Minimum of Square Error), assigned to the mean value T1, the next macroblock as other macroblocks threshold value: if the decision is only 14 mode, if the T1 constant of 14 and 116 sequentially mode decision, the MSE mean both assigned to Tl. combine the above methods of the present invention, the intra prediction mode selection. the results showed that the coding speed can be increased several times, the coding rate and to ensure the full search and PSNR compared little change. Figure 1 of the present invention is an information processing block diagram of a video encoding embodiment. FIG. 2 is an invention 实施例的处理流程图图3是4x4块帧内预测方向图图4是当前块的邻块位置图图5是116和14的率失真变化图具体实施方式以下结合附图描述本发明的H.264/AVC实时编码器的帧内预测模式的快速选择方法的一个优选实施例-在校验模型加72编码器上,具体工作环境为2. 4GHz CPU的PC机,针对I帧序列,编码序列为IIII,熵编码采用了CABAC,量化参数分别为14、 28,亮度14的RDO 阈值T0和宏块复杂度的阈值T1可自适应计算得到,采用下述方法:在我们实验过程中,发现校验模型提供的全搜索方法忽略了模式选择间的相关性。在实际中,宏块的色度块和亮度块帧内预测模式间、4x4亮度块(14)和16x16亮度块(116)的模式选择之间以及亮度块预测模式内部都具有较强的相关性。如果能充分利用这些相关性,就可以快速有效的确定帧内预测的最佳模式。(1)参见图1和图2,首先,可以观察 A process flow diagram of an embodiment 3 is a block intra 4x4 prediction direction in FIG. 4 is a block ortho position of a current block of FIG. 5 of the present invention is described with reference to a distortion rate changes 16 and 14 below in connection with FIG H. DETAILED DESCRIPTION a rapid method of selecting an intra prediction mode 264 / AVC preferred embodiment real-time encoder Example - 72 plus parity encoder in the model, the specific working environment of the PC 2. 4GHz CPU, the I-frame sequence, the coding sequence for a as IIII, entropy encoding using the CABAC, the quantization parameter, respectively 14, 28, RDO threshold T0 luminance 14 and macroblock complexity threshold T1 may be adaptively calculated using the following method: in our experiment, we found the check the method of providing full search model ignores the correlation between mode selection. in practice, between the luminance blocks and chrominance blocks intra prediction mode of a macroblock, 4x4 luma block (14) and a 16x16 luminance block (116) a mode selection and an internal brightness between the prediction mode of block have a strong correlation. If we can take advantage of these dependencies, can quickly and efficiently determine the best intra prediction mode. (1) Referring to Figures 1 and 2, first of all, can be observed 到亮度宏块无论选择I16还是I4,色度宏块模式都是一致的,因此可以确定在整个模式选择过程中色度块的模式选择具有独立性,故色度预测模式可以在亮度判决之前进行独立判决色度块的最佳预测模式可以为亮度宏块模式判决提供依据。尽管对于亮度I16与色度18模式选择不能达到完全一致,但存在着联系:当色度块选择D模式,亮度可以选择D、 H、 V、 P模式;当色度块选择H模式,亮度可以选择D、 H、 P模式;当色度块选择V模式,亮度可以选择D、 V、 P模式;当色度块选择P模式,亮度可以选择D、 P模式;而对于亮度I4与色度18, 18选择的模式是亮度I4下绝大多数4x4块最有可能选择的模式。 I16 select whether or luminance I4, chroma macroblock macroblock mode are the same, so the process of determining the mode of the chroma blocks may be selected across the mode selection independent, so the chroma prediction mode may be performed before the brightness judgment the best prediction mode decision independently chrominance block may provide a basis for the luminance macroblock mode decision although the luminance and chrominance I16 mode selector 18 can not reach exactly the same, but there is a link: when chroma D block selection mode, brightness can be select D, H, V, P mode; H block selection mode when the chroma, the brightness can choose D, H, P mode; when the V chrominance block mode selected, the brightness can choose D, V, P mode; when the chrominance blocks P mode is selected, the brightness can choose D, P mode; I4 and the luminance and chrominance 18, 18 at a luminance of the selected mode is the vast majority of 4x4 blocks I4 most likely mode. 亮度块内部模式选择之间存在着优先性,V、 H、 D模式相对于其它模式具有优先选择性。 Exists between luminance priority mode selection block inside, V, H, D mode has priority with respect to other modes selectively. 14 情况下,通过相邻块(图4)的模式可以判决得到当前块的最优先选择模式。 14 the case, (FIG. 4) may decide the mode of the neighboring block to obtain the most preferred mode of the current block. 对于优先选取的模式,其方向上相邻的模式也具有很大的选中可能。 Priority for the selected mode, which is also adjacent in the direction pattern may be selected with great. 实际中,14的选择概率远远大于116,因此在判决过程上优先检测14。 In practice, the selection probability is far greater than 14 116, 14 thus detected on the priority decision process. 在I4情况下在确定的16个4x4块最佳模式中,概率最大的两种模式很可能与116的最佳模式一致,因此可以将之色度模式联合进行检验,最终确定I16最佳模式。 In the 4x4 block 16 determines the best mode, the two modes is the maximum probability may coincide with the best mode at 116 I4 case, the chroma mode can thus be combined inspection, I16 ultimately determine the best mode. H.264/AVC编码器采用RD0为模式判别准则,但其结构复杂,尤其写码流相当耗时间,因此可以根据具体模式采用不同的判决依据:①色度模式选择不依赖亮度的模式选择,且其编码比特只包含色度模式和色度残差两部分,因此采用SAD就可以反映不同模式间率失真的差别;②在亮度14的4x4块编码比特包括块模式和残差的熵编码。 H.264 / AVC encoder uses RD0 criterion as a model, but the complexity of its structure, in particular relatively time stream writing, can be different according to judgment based on the specific mode: ① chroma mode selection does not depend on the brightness mode selection, and the coded bits which contains only two residual chroma mode and a chrominance portion, so using SAD may reflect differences between the different modes of distortion; ② mode and a residual block comprises entropy coding 4x4 blocks in the coded bit luminance 14. 因为4X4块尺寸较小,其块模式编码的比特数和不同模式间的残差编码比特数的差异会导致率失真的变化很大,故选择率失真作为14的l莫式判决依据。 4X4 because of the smaller block size, the number of residual differences between the coded bits coded block pattern whose number of bits and different modes of distortion can lead to large changes, so as to select a rate distortion l 14 Mohs judgment basis. 且4x4块的率失真的分布比较集中,可在分布范围中定义一个阈值T0,如果在某一模式下其率失真小于T0,可认为达到最佳模式;③116编码比特主要包括宏块模式和残差的熵编码,故SAD能够反映率失真的变化,以其作为模式选择的依据。 And distortion of 4x4 blocks concentrated distribution, can be defined in a distribution range of the threshold value T0, if a mode in which the rate-distortion is less than T0, can be considered the best mode; ③116 encoded bits macroblock mode and includes residues poor entropy coding, so the SAD to reflect changes in the rate of distortion, as the basis for its selected mode. 由于I16和I4随宏块纹理复杂度的RD0曲线是不同的(图5),在模式选择时还应考虑宏块的纹理复杂度,并以宏块纹理复杂度的某个值作为阈值T1来确定二者间的判决,即在纹理平坦区域(小于T1), 116生成的重建图像能更好地与原始图像相匹配,选择I16概率大于I4,随着纹理复杂度的逐渐增加,116在细节内容上率失真逐渐增加,当大于或等于T1时将大于I4的率失真,这时14更适合宏块编码模式选择。 Since I16 and I4 with macroblock RD0 Texture Complexity profile is different (FIG. 5), should also be considered when the mode selection texture complexity macroblock, a macroblock and the texture complexity of a value as a threshold value T1 to determining the decision between the two, i.e., a flat area in the texture (less than Tl), the generated reconstructed image 116 to better match the original image, I4 selection probability is greater than I16, with the increasing complexity of texture, detail 116 rate distortion content is gradually increased, when T1 is equal to or greater than the rate of greater than I4 artifact, and 14 is more suitable for macroblock coding mode selection. 衡量纹理复杂性的方式有很多种,如算术平均、几何平均、中值、方差和均方误差。 There are many ways to measure the complexity of texture, such as arithmetic mean, geometric mean, median, variance and the mean square error. 本发明选用了均方误差衡量宏块的纹理复杂度。 The present invention selects the macro block error is a measure of mean square texture complexity. 本发明充分利用上述的相关性,将之有机结合在一起,根据RDO准则确定不同模式的判决依据,得到帧内预测模式的快速选择方法,其流程如图2所示,具体步骤实施如下:① 对当前宏块的色度块进行帧内预测,以SAD为判决依据,确定色度宏块帧内的最佳模式。 The present invention fully utilizes the above-described correlation, the combination of an organic, depending on the decision is determined in accordance with RDO mode, rapid method of selecting an intra prediction mode, the process shown in Figure 2, the steps of implementation as follows: ① chroma blocks of the current intra-prediction macroblock, as a decision based on SAD to determine the best mode of the chroma macroblock frame. ② 对当前宏块的亮度块进行I4帧内预测,首先根据当前4x4块的相邻块(图2) 的模式确定可能模式,将之和色度块的最佳模式相比较,确定最有可能块模式。 ② the current macroblock luminance block I4 intra prediction mode may be determined according to the first neighboring blocks of the current 4x4 blocks (FIG. 2) mode, and the best mode of the chrominance blocks compared to determine the most likely block mode. 计算最可能模式的J^^,与阈值T0比较,如果小于T0,即可确认为最佳模式,跳转到步骤⑦。 To calculate the most likely mode J ^^, T0 is compared with a threshold value, is less than T0, can be recognized as the best mode, jumps to step ⑦. 否则跳转到步骤(D。③ 计算另一模式(上面被排除的模式)的《^。。£,与步骤②计算的J^^比较,得到最小J说M值的模式。与T0比较,如果小于TO,确认为最佳模式,跳转到步骤⑦。④ 选择最小J^^的模式作为候选模式,计算与其预测方向相邻的两个模式的,比较这三种模式的A,,如果邻近模式的J,£值都大于候选模式的Jmpw,跳转步骤⑦。否则到步骤(D。 ◎如果最小J^^的模式是V、 H或D,计算这三种模式在上述过程未曾计算的4,,比较得到最小4,的模式,跳转到步骤⑦;如果是其它模式,跳转到步骤⑥。©计算其它剩余模式,得到最小J^^模式,跳转到步骤⑦。⑦ 该最小J^^模式为该4x4块的最佳模式。⑧ 循环步骤②一⑦,得到16个4X4块的最佳模式。计算宏块复杂度,如果大于阈值T1 (图3),则该宏块为I4,结束判决。否则计算14情况下整个宏块的随后,在步骤24 Otherwise jumps to step (D.③ another calculation mode (mode above is excluded) is "^ .. £, calculated in step ② J ^^ comparing said pattern to obtain the minimum value M J Compared with T0, If less than the tO, recognized as the best mode, jumps to step ⑦.④ J ^^ select the smallest candidate mode as the mode to calculate its two modes of adjacent prediction directions, a comparison of these three modes, that if J mode adjacent, £ Jmpw candidate value is greater than the mode jump step ⑦. otherwise, the step (D. ◎ if the pattern is the minimum J ^^ V, H or D, which is calculated in the above three modes never calculation process comparing the obtained minimum 4 ,, 4, the mode jumps to step ⑦;. If the other mode, jumps to step ⑥ © other remaining mode is calculated to give the minimum J ^^ mode, jumps to the step ⑦.⑦ minimum J ^^ mode best mode for 4x4 blocks .⑧ cycle ② a step ⑦, 16 give the best mode 4X4 block. macroblock complexity calculation, if greater than the threshold T1 (FIG. 3), the macroblock is I4, the judgment ends. otherwise, the entire macroblock is calculated and then the case 14, at step 24 ,给定a,的搜索步长△, =0.05 ,并令a】=—6.0 + 0.05/n2 = —6.0 ;随后,在步骤25中计算出当前a。,^状况下的各个钢巻的轧制压力值{/^} = {142345.2,165215.8,659984.7,193181.3,252650.4,117565.3,922542.5, 105623.5,881502.1,167711.1,157397.6,125747.9,724300.4,1373038,652179.6, 153379.9,813176.6,251275.7,217451.5,3,61E+07,48160.95,567036.6,685285.3, 91783.1,82468.63,385802.3,148810.4,225057,71380.36,70728.84,72383.42, 166550.4,261870.8,206352.7,157700.8}随后,在步骤26中计算当前状况下目标函数的具体数值随后,在步骤27中定义初始最佳钢种与工况影响系数为"^,^,与之对应的轧制压力计算值为尸;,判断不等式巧<尸。是否成立?显然成立,此时令F0 =巧=1.03xl06 , 〜=fl0=-10 , 6.0 ,{尸、} = {/^} = {142345.2,165215.8,659984.7,193181.3,252650.4,117565.3,922542.5, 105623.5,881502.1,167711.1,157397.6,125747.9,724300.4,1373038,652179.6, 153379.9,813176.6,251 , A given, the search step △, = 0.05, and let a] = - 6.0 + 0.05 / n2 = -6.0; then calculate the current in a step 25, each steel ^ Volume under rolling conditions. braking pressure value {/ ^} = {142345.2,165215.8,659984.7,193181.3,252650.4,117565.3,922542.5, 105623.5,881502.1,167711.1,157397.6,125747.9,724300.4,1373038,652179.6, 153379.9,813176.6,251275.7,217451.5,3,61E +07,48160.95,567036.6,685285.3, 91783.1,82468.63,385802.3,148810.4,225057,71380.36,70728.84,72383.42, 166550.4,261870.8,206352.7,157700.8} then, the objective function is calculated under the current conditions in step 26 and then specific value, in step 27 the initial definition of the optimum conditions of steel with a coefficient of influence "^, ^, corresponding calculated value of the pressure rolling dead; determining inequality coincidence <clearly established corpse is satisfied, so at this time F0 =.? Qiao = 1.03xl06, ~ = fl0 = -10, 6.0, {corpse,} = {/ ^} = {142345.2,165215.8,659984.7,193181.3,252650.4,117565.3,922542.5, 105623.5,881502.1,167711.1,157397.6,125747.9,724300.4 , 1373038,652179.6, 153379.9,813176.6,251 275.7,217451.5,3.61E+07,48160.95,567036.6,685285.3, m2 = m2 +1 , 91783.1,82468.63,385802.3,148810.4,225057,71380.36,70728.84,72383.42, 166550.4,261870.8,206352.7,157700.8}转入步骤28。 275.7,217451.5,3.61E + 07,48160.95,567036.6,685285.3, m2 = m2 +1, 91783.1,82468.63,385802.3,148810.4,225057,71380.36,70728.84,72383.42, 166550.4,261870.8,206352.7,157700.8} proceeds to step 28. 如果不成立,则令附2=附2+1后转入步骤28;随后,在步骤28中,判断不等式附2^12/^=240是否成立,如果成立, 则转入步骤24。 If false, then let the attached attachment 2 = 2 + 1 proceeds to step 28; then, in step 28, it is judged attached inequality 2 ^ 12 / ^ = 240 is established or if established, the process proceeds to step 24. 否则,令附1=附,+1转入步骤29;随后,在步骤29中,判断不等式m^20/A。 Otherwise, let 1 = attachment attached, + 1 proceeds to step 29; then, in step 29, it is determined inequality m ^ 20 / A. -400是否成立,如果成立,则转入步骤23,否则转入步骤30;随后,在步骤30中,输出初始最佳钢种与工况影响系数为"^=-1.95, ~= -5.85 以及与之对应的轧制压力计算值为{尸;,} = {224,315,282,306,314,309,311,347,271 ,247,267,231,279,302,303,240,302,264, 307,257,350,304,302,292,283,268,304,215,294,301,280,295,302,278,300,} '随后,在步骤31中,定义g为实际轧制压力f与。。"^所对应的轧制压力计算值尸,,差的绝对值, 即取={53.1049,80.2467,41.3122,60.2268,61.243,46.756,28.7658,114.4855,0.1316,32.5427,30.5383,23.7743,31.2256,19.743,12.0435,26.0086,5.5041,17.8542,33.4429,214.4543,92.3994,12.2874,2.38931.2688,16.5762,0.7952,15.4913,1.3435,4.0772,14.8903,5.8575,27.8979,10.5256,20.4002,21.0803} 残差矩阵可视为0矩阵。 -400 is established or if established, the process proceeds to step 23, otherwise goes to step 30; then, in step 30, the initial output and the optimum operating conditions of steel impact coefficient "^ = - 1.95, and -5.85 ~ = the corresponding rolling pressure calcd {corpse;,} = {224,315,282,306,314,309,311,347,271, 247,267,231,279,302,303,240,302,264, 307,257,350,304,302,292,283,268,304,215,294,301,280,295,302,278,300,} 'subsequently, in step 31, g is defined as f and the actual rolling pressure .. "^ corresponding rolling pressure calcd dead ,, the absolute value of the difference, that is to take = {53.1049,80.2467,41.3122,60.2268,61.243,46.756,28.7658,114.4855,0.1316 , 32.5427,30.5383,23.7743,31.2256,19.743,12.0435,26.0086,5.5041,17.8542,33.4429,214.4543,92.3994,12.2874,2.38931.2688,16.5762,0.7952,15.4913,1.3435,4.0772,14.8903,5.8575,27.8979,10.5256,20.4002 , 21.0803} residue matrix can be regarded as a zero matrix. 即满足下面的不等式:I DC |< (29-to —砂—co"rf)〜"朋Lcoe/[砂—rem][O][O] 公式(5) 其中: 砂—per = g尸/ 6 , "附=2尸%6 , 砂—+ 15 ,W_co"W = (l 6 , QP为量化参数。对于每个4x4块,公式(6)给出SAD的计算,具体如下:&4" = ^^|^0,>0-?(x,力l 公式(6)—很明显下面不等式成立:3 3 3 3i Z 2 oo,力—》'o,力)卜2] X!(》o,力—-' <>,力i;c=0少=0 x=0 7=0艮卩IDCg&4仏则如果下式成立,则满足O矩阵的判断。&4£»<(29-te-砂—co"0〜w朋f — coe/[砂—re附][0][0] 公式(7)阔值TO由4x4块的RD0确定,如公式8所示,mode表示所选的预测模式,i?包括模式类型和残差的熵编码。^^(mode^^D + z^d^Cmode^) 公式(8)其中,&Si) = t^>0c,3;)-p(x,力l , p[jc,xl是预测后残差经过变换、量化、反量化、:c=0 "0反变换后与参考块的值相加得到,故下面不等式成立:i《0,力—P(x,力|<| ;?(x, j;) - ^ (x,;/) | 因此柳S (过| s(x,力|2) < (过| s I.e., satisfies the following inequality: I DC | <(29-to - Sand -co "rf) ~" Peng LCOE / [sand -rem] [O] [O] Equation (5) wherein: sand dead -per = g / 6, "attachment 2 = 6% dead, sand - + 15, W_co" W = (l 6, QP is a quantization parameter gives the SAD is calculated for each 4x4 block, equation (6), as follows: & 4 "=. ^^ | ^ 0,> 0 - (x, l force equation (6) - it is clear that the following inequality: 3 3 3 3i Z 2 oo, force -? " 'o, force)! Bu 2] X (" o force - '<>, power i; c = 0 less = 0 x = 0 7 = 0 Gen Jie IDCg & 4 Fo then if the following holds, is satisfied O matrix Analyzing & 4 £ »<(29-te- sand. -co "0~w Points f - coe / [-re attachment sand] [0] [0] equation (7) is determined by the width value tO RD0 4x4 block, as shown in equation, mode indicates the selected prediction mode 8 , i and comprises a mode type of entropy coding the residual ^^ (mode ^^ D + z ^ d ^ cmode ^) equation (8) where, & Si) = t ^> 0c, 3;?.) - p (x, force l, p [jc, xl transformed prediction residuals after quantization, inverse quantization,: c = 0 "0 after the inverse transform value obtained by adding the reference block, so that the following inequality holds: i" 0, force - P (x, force | <|; (x, j;) - ^ (x,;? /) | Thus Liu S (through | s (x, force | 2) <(Guo | S (x,力|)2x=0 "0 义=0 "0故5SD〈&4D2,艮卩SS£> < ((29 —砂—co"W) /《wa对_ coe/[gp — rem][0][O])2 公式(9 ) 这时,该块残差量化系数将趋于0,因此得到TO。T0 = ((2?-te —砂—co"0〜waW一coe/[砂—rem][0][0])2 公式(10)实际中可能存在差别,因此设置一个校正因子《 = 0.5:<formula>formula see original document page 12</formula> 公式(11)(3)计算阈值T1的自适应方法在实际中,当T1越大,搜索I16的可能性越大,因此搜索的模式越多;Tl越小, 搜索116的可能性越小,则搜索的模式变少。本发明通过当前宏块的前一宏块来确定当前T1的值,具体方法如下:初始化:Tl=l《20第一个宏块:依次进行I4和I16模式判决,得到二者的MSE,将均值赋给T1, 作为下一宏块的阈值。其它宏块:如果仅判决了14模式,则Tl不变。如果依次进行了14和116模式判决,则二者的MSE的均值赋给T1。 (X, force |) 2x = 0 "0 Yi = 0" 0 so 5SD <& 4D2, Gen Jie SS £> <((29 - Sand -co "W) /" wa pair _ coe / [gp - rem] [ 0] [O]) 2 equation (9) In this case, the residual block is quantized coefficients will tend to 0, thus obtaining TO.T0 = ((2 -te -? sand -co "0~waW a coe / [sand ] [0] [0]) 2 equation (10) may be present in the actual difference -rem, thus setting a correction factor "= 0.5: <formula> formula see original document page 12 </ formula> equation (11) (3) calculating the threshold value T1 adaptive approach in practice, when the T1, the greater the possibility of searching I16, so the more the search pattern; Tl, the smaller the possibility of searching 116, the search pattern is reduced. the present invention is determined by the current value of the current T1 previous macroblocks, the specific method is as follows: initialization: Tl = l "20 first macroblocks: sequentially I16 and I4 mode decision, MSE obtained both the mean assigned Tl, as the threshold value of the next macroblock other macroblocks: only if the decision model 14 is sequentially performed Tl unchanged if the mode decision 14 and 116, the MSE mean both assigned to Tl.. 测试三个纹理复杂程度不同的CIF序列Mobile、 Bus和Foreman。 Testing three different textures of the complexity of the CIF sequence Mobile, Bus and Foreman. 实验结果表明, 本快速帧内预测模式选择方法比H. 264/AVC原有的全搜索方法可提高I帧编码速度大约6倍左右。 Experimental results show that the present method for fast intra prediction mode selection than H. 264 / AVC original full-search method can improve the rate of about I frame encoding is about 6 times. 在速度提高的同时,平均信噪比变化约O. ldb左右,码率稍有变化。 While improving the speed, average SNR variation around about O. LDB, a slight change rate. 本发明的第二个实施例是:针对P帧编码,编码序列为I PPP ,运动估计搜索范围±16, 5个参考帧,运动估计算法为全搜索算法,熵编码采用了CABAC,量化参数分别为14、 28。 A second embodiment of the present invention is: for a P-frame coding, the coding sequence for the I PPP, motion estimation search range of ± 16, 5 reference frames, the motion estimation algorithm is a full search algorithm, using the CABAC entropy coding, the quantization parameter, respectively 14, 28. 亮度14 4x4块的RD0的阈值T0和宏块复杂度的阈值Tl,同样采用上述的自适应方法。 RD0 14 4x4 luminance block thresholds Tl T0 and macroblock complexity, adaptive method described above using the same. 仍然测试三CIF序列Mobile、 Bus和Foreman。 Still testing three CIF sequence Mobile, Bus and Foreman. 实验结果表明, 本发明的快速帧内预测模式选择方法在帧间情况下比H.264/AVC原有的帧间全搜索方法可提高编码速度大约25°/。 Experimental results show that the present invention is fast intra mode prediction method selection ratio H.264 / AVC original full-search method can be improved inter encoding speed in the case of an inter about 25 ° /. 左右。 about. 在速度提高的同时,平均信噪比变化约0.02db 左右,码率几乎没有变化。 While improving the speed, average SNR variation of about 0.02dB around, almost no change rate.

Claims (1)

1. 一种H.264/AVC实时编码器帧内预测模式的快速选择方法,在H.264/AVC编码器中,帧内编码采用帧内预测方法,其特征在于利用当前宏块的色度块和亮度块帧内预测模式间的相关性、4x4亮度块I4和16x16亮度块I16的模式选择之间的相关性以及亮度块预测模式内部的相关性,以简化的率失真优化RDO为判决依据,快速进行帧内预测模式选择;快速选择的步骤如下: a.确定色度块的模式选择具有独立性,即亮度宏块无论选择I16还是I4,色度宏块模式都是一致的,从而首先对色度预测模式以SAD为准则进行独立判决,确定色度的最佳预测模式; b.确定了色度的最佳预测模式后,为亮度快模式判决提供如下依据:对于亮度I16而言,当色度块选择直接D模式,亮度可以选择直接D、水平H、垂直V、平坦P模式; 当色度块选择H模式,亮度可以选择直接D、水平H、平坦P模式;当色 Quick selection method for H.264 / AVC real time encoder intra prediction mode in H.264 / AVC encoder, intra-coded intra-frame prediction method, wherein the chrominance of the current macroblock using correlation correlation and internal brightness prediction mode of block selecting between correlation between the block and the luminance block intra prediction modes, 4x4 luminance block of 16x16 luma blocks I16 and I4 mode, a simplified rate-distortion optimization RDO decision is based , fast intra prediction mode selection; quick selection step as follows:. a determination mode selecting chroma blocks having independent, i.e., whether selection I16 luminance macroblock or I4, chroma macroblock mode are the same, so that firstly for chroma prediction mode for the SAD criterion independent decision to determine the best prediction mode, chroma; b after determining the optimum prediction mode chroma, brightness mode provides quick decision based on the following: the luminance I16 concerned, when chroma block D selected direct mode, the brightness can directly select the D, horizontal H, vertical V, P planar pattern; H block selection mode when the chroma, the brightness can directly select the D, H level, planar model P; when the color 度块选择垂直V模式,亮度可以选择直流D、垂直V、平坦P模式;当色度块选择平坦P模式,亮度可以选择直流D、平坦P模式;而对于亮度块I4,亮度块I8选择的模式是亮度块I4下绝大多数4x4块可能选择的模式; c.对于宏块,I4的选择概率远远大于I16,因此我们首先进行I4下4x4块的模式判决;根据邻块的模式判断得到当前块的可能模式,并和由步骤b确定的可能模式,分别计算这两种模式下的RDO,并与阈值T0比较,小于T0,则可确定为当前4x4的最佳模式,否则较小值RDO的为可能最佳模式; d.比较步骤c中得到的可能最佳模式的预测方向邻近的模式的RDO,进一步确定可能的最佳模式。 V of the vertical block selection mode, DC D may be selected luminance, V the vertical, planar model P; P when the flat chrominance block selection mode, DC D may be selected brightness, the P mode is flat; for a luminance block I4, I8 selected luminance blocks mode is the vast majority of 4x4 luminance block I4 may block mode selected;. c for the macroblock, the selection probability is far greater than I16 I4, I4 we first performed at decision block 4x4 mode; neighbor blocks is determined according to a mode to give possible mode of the current block, and step b, and the determination of possible patterns, calculate RDO in both modes, and T0 is compared with a threshold value, less than T0, it may be determined as the current best mode of a 4x4 or smaller value It may be the best mode for RDO; d RDO prediction direction may be the best mode of comparison obtained in step c of an adjacent pattern, and further to determine the best possible mode. 如果它是垂直V、水平H或直流D,则比较这三种模式中未曾比较的模式;如果它为其它模式,则比较所有其它未曾比较的模式; e.在完成当前宏块的16个4x4块的模式判决后,计算宏块纹理复杂度,以MSE来衡量,并与阈值T1比较,如果大于T1,提前终止判决,上述得到的结果为宏块最佳模式;如果小于T1,则进行亮度16x16块的模式判决; f.根据步骤b中色度块的最佳模式与亮度块的关系,确定当前宏块I16的可能模式;统计步骤e中I4情况下16个子块的模式,得到概率最大的两种模式;综合考虑这些可能模式,最终确定I16的可能模式;并以SAD为准则,确定I16的最佳预测模式; g.自适应确定步骤e所述的阈值T1来判断是否提前终止I16判决和自适应确定步骤c所述的T0来判断是否提前终止I4的判决; 所述的阈值通过自适应方法获取的具体步骤为: 阈值T0通过如下公式自适 If it is the vertical V, horizontal H or DC D, the comparison of the three modes never comparison mode; if it is the other mode, the comparator has not been compared to other modes all; the completion of the current macro block E 16 4x4. after the mode of the block decision, calculated macroblock texture complexity to MSE measured, and T1 is compared with a threshold value, if greater than T1, early termination judgment result of the obtained macroblock best mode; is less than T1, the luminance 16x16 block mode decision;. f best mode according to the relationship in step b luminance block and chrominance block, a possible mode of the current macroblock I16; in the case of the statistical model in step e I4 16 sub-blocks, the probability of getting the maximum the two modes; considering these possible modes, I16 finalized possible modes; SAD criteria and to determine the optimal prediction mode I16;. g adaptively determined according to step e to determine whether the threshold value T1 early termination I16 adaptively determining decision and said step c to determine whether to terminate the judgment T0 I4 in advance; the specific steps taken by an adaptive threshold method: adaptive threshold value T0 by the following formula 应获取: T0=α×((2q-bits-qp_const)/quant_coef[qp_rem][0][0])2 其中,α是校正因子,其取值范围在(0,1],qp_rem=QP%6,qp_bits=qp_per+15,qp_const=(1<<q_bits)/6,quant_coef是量化矩阵,QP为量化参数; 计算阈值T1的自适应方法为: 初始化T1=1<<20; 第一个宏块:依次进行I4和I16模式判决,得到二者的MSE(Minimum ofSquare Error),将均值赋给T1,作为下一宏块的阈值。 其它宏块:如果仅判决了I4模式,则T1不变。如果依次进行了I4和I16模式判决,则二者的MSE的均值赋给T1。 Should be obtained: T0 = α × ((2q-bits-qp_const) / quant_coef [qp_rem] [0] [0]) 2 where, [alpha] is a correction factor, which is in the range (0,1], qp_rem = QP% 6, qp_bits = qp_per + 15, qp_const = (1 << q_bits) / 6, quant_coef quantization matrix, QP is a quantization parameter; adaptive method for calculating the threshold value T1: initialization T1 = 1 << 20; first macro block: sequentially mode decision I16 and I4, both obtained MSE (Minimum ofSquare Error), assigned to the mean value T1, the threshold value as a next macroblock other macroblocks: if the decision I4 only mode, the constant T1 If in turn the mode decision I16 and I4, the MSE mean both assigned to Tl.
CN 200410025123 2004-06-11 2004-06-11 Quick selection of prediction modes in H.264/AVC frame CN100401789C (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN 200410025123 CN100401789C (en) 2004-06-11 2004-06-11 Quick selection of prediction modes in H.264/AVC frame

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN 200410025123 CN100401789C (en) 2004-06-11 2004-06-11 Quick selection of prediction modes in H.264/AVC frame

Publications (2)

Publication Number Publication Date
CN1585495A CN1585495A (en) 2005-02-23
CN100401789C true CN100401789C (en) 2008-07-09

Family

ID=34601155

Family Applications (1)

Application Number Title Priority Date Filing Date
CN 200410025123 CN100401789C (en) 2004-06-11 2004-06-11 Quick selection of prediction modes in H.264/AVC frame

Country Status (1)

Country Link
CN (1) CN100401789C (en)

Families Citing this family (30)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN100413342C (en) 2005-03-09 2008-08-20 浙江大学 Method and device of frame prefiction mode coding decoding for video or image compression
CN100484249C (en) 2005-09-25 2009-04-29 海信集团有限公司 Method of forecast inside frame based on edge direction between frames
CN1859576A (en) 2005-10-11 2006-11-08 华为技术有限公司 Top sampling method and its system for space layered coding video image
JP5089878B2 (en) 2005-10-28 2012-12-05 パナソニック株式会社 Image encoding device
CN100417228C (en) 2005-10-31 2008-09-03 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
CN100409690C (en) 2006-05-19 2008-08-06 北京工业大学 Video interframe compression method based on space-time correlation
WO2008056923A1 (en) 2006-11-07 2008-05-15 Samsung Electronics Co, . Ltd. Method and apparatus for encoding and decoding based on intra prediction
CN101198051B (en) 2006-12-07 2011-10-05 深圳艾科创新微电子有限公司 Method and device for implementing entropy decoder based on H.264
CN100508610C (en) 2007-02-02 2009-07-01 清华大学 Method for quick estimating rate and distortion in H.264/AVC video coding
CN100551075C (en) 2007-10-15 2009-10-14 中兴通讯股份有限公司 Low-complexity in-frame estimation mode selection method
CN100574460C (en) 2008-01-31 2009-12-23 上海广电(集团)有限公司中央研究院 AVS intra-frame prediction reference pattern extracting method
CN101521819B (en) * 2008-02-27 2010-12-01 深圳市融创天下科技发展有限公司 Method for optimizing rate distortion in video image compression
CN100586185C (en) 2008-04-10 2010-01-27 清华大学 Mode selection method for transcoding 264 video to reduce resolving capability
CN101340581B (en) 2008-08-15 2010-06-02 上海富瀚微电子有限公司 Intra-frame prediction mode selection method and its apparatus based on M algorithm
PL2465265T3 (en) * 2009-08-12 2019-07-31 Interdigital Vc Holdings, Inc. Methods and apparatus for improved intra chroma encoding and decoding
JP4851580B2 (en) 2009-11-20 2012-01-11 シャープ株式会社 Image reading apparatus, image data output processing apparatus including the same, and image reading method
US8306343B2 (en) * 2010-04-09 2012-11-06 Newport Media, Inc. Optimized prediction based image compression
CN101888547B (en) * 2010-06-18 2012-12-05 西安电子科技大学 Direct mode selection based H.264/AVC (Advanced Video Coding) fast transcoding method and device
CN102469308B (en) * 2010-11-15 2015-05-13 深圳市云宙多媒体技术有限公司 Video coding intra-frame prediction method
CN102055980B (en) * 2010-12-31 2013-03-20 北京大学深圳研究生院 Intra-frame predicting circuit for video coder and realizing method thereof
JP5636507B2 (en) * 2011-01-07 2014-12-03 メディア テック シンガポール ピーティーイー.リミテッド Method and apparatus for improved intra prediction mode coding
CN102186070B (en) * 2011-04-20 2013-06-05 北京工业大学 Method for realizing rapid video coding by adopting hierarchical structure anticipation
KR101753551B1 (en) * 2011-06-20 2017-07-03 가부시키가이샤 제이브이씨 켄우드 Image encoding device, image encoding method and recording medium storing image encoding program
EP2919467A3 (en) 2011-06-28 2015-10-14 Samsung Electronics Co., Ltd Method and apparatus for image encoding and decoding using intra prediction
KR20130049522A (en) * 2011-11-04 2013-05-14 오수미 Method for generating intra prediction block
EP2859724B1 (en) 2012-06-22 2019-09-04 MediaTek Inc. Method and apparatus of adaptive intra prediction for inter-layer coding
CN102740077B (en) * 2012-07-04 2014-10-01 西安电子科技大学 Selection method based on an intra prediction mode h.264 / avc standard
CN102946538B (en) * 2012-11-09 2015-03-25 中国矿业大学 Rapid intra-frame mode selection method suitable for downhole distribution type video encoding
CN103096090B (en) * 2013-02-20 2015-12-02 广州柯维新数码科技有限公司 A method for video compression coding block divided for
CN103581682B (en) * 2013-08-14 2017-08-01 北京交通大学 A kind of fast mode decision algorithm of HEVC intraframe codings and its application

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1336080A (en) 1999-10-29 2002-02-13 皇家菲利浦电子有限公司 Video encoding-method
CN1492687A (en) 2003-09-04 2004-04-28 上海大学 Quick full pixel movement evaluation method of multiple kinds of modules

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1336080A (en) 1999-10-29 2002-02-13 皇家菲利浦电子有限公司 Video encoding-method
CN1492687A (en) 2003-09-04 2004-04-28 上海大学 Quick full pixel movement evaluation method of multiple kinds of modules

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
H.264变换和量化的分析. 楼剑,陆亮,虞露,董洁.浙江大学学报,第38卷第5期. 2004
Rate-Constrained Coder Control and Comparison ofVideo Coding Standards. Thomas Wiegand, Heiko Schwarz,AnthonyJoch,FaouziKossentini.IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY,Vol.13 No.7. 2003
一种基于H.264/AVC的高效块匹配搜索算法. 薛金柱,沈兰荪.电子学报,第32卷第4期. 2004

Also Published As

Publication number Publication date
CN1585495A (en) 2005-02-23

Similar Documents

Publication Publication Date Title
CN1143551C (en) Device for predicting and encoding image
CN101208873B (en) Rate control for multi-layer video design
CN1981535B (en) Method and apparatus for weighted prediction in predictive frames
JP3964391B2 (en) Image encoding apparatus, image decoding apparatus, image encoding method, and image decoding method
KR100719647B1 (en) Video data conversion device and video data conversion method
CN102084655B (en) Video encoding by filter selection
CN101502120B (en) Adaptive geometric partitioning method and device for video decoding
CN101099394B (en) Method and apparatus for fast mode decision of b-frames in a video encoder
CN101416524B (en) Dynamic selection of motion estimation search ranges and extended motion vector ranges
CN101273641B (en) Dynamic image encoding device and dynamic image decoding device
JP2006513636A (en) Fast mode decision making for inter-frame coding
CN101494782B (en) Video encoding method and apparatus, and video decoding method and apparatus
CN1206864C (en) Association rate distortion optimized code rate control method and apparatus thereof
CN101946515B (en) Two pass quantization for cabac coders
CN101448162B (en) Method for processing video image
CN103179395B (en) A rate of context adaptive variable length coding (CAVLC) - the quantization distortion
CN100568976C (en) Method and device for intra prediction coding and decoding of image
KR100627329B1 (en) Apparatus and method for adaptive motion estimation and mode decision in h.264 video codec
CN100566426C (en) Method and apparatus for encoding and decoding of video
CN100534194C (en) Methods and apparatus for video intraprediction encoding and decoding
CN1316830C (en) Moving picture coding apparatus, moving picture decoding apparatus, moving picture coding method, and moving picture decoding method
CN101889405B (en) Method and apparatus for performing motion estimation
CN101529916B (en) Video encoding with intra encoding selection
CN102098517B (en) Method and apparatus for determining expected distortion in decoded video blocks
CN102939761B (en) A method and apparatus for using an adaptive filter to perform intra prediction

Legal Events

Date Code Title Description
C06 Publication
C10 Request of examination as to substance
C14 Granted
C17 Cessation of patent right