Now address:Home >> newlist.asp >> Company News >> news content
H.264 Technology
Send time:2014-07-10 Popularity:

  1,Hierarchical design

  H. 264 algorithm in concept can be divided into two layers: video coding layer (VCL: VideoCodingLayer) is responsible for the efficient video content, said network abstraction layer (NAL: NetworkAbstractionLayer) is responsible for the appropriate required by network way for packaging and transmit data. Between the VCL and NAL defines an interface based on grouping, packaging and the corresponding signal to a part of the NAL. In this way, high coding efficiency and network friendliness tasks are done by the VCL and NAL respectively.

  The VCL layer including the block-based motion compensation hybrid coding and some new features. As in the previous video coding standard, h. 264 didn't pre and after processing functions such as included in the draft, so that we can increase the flexibility of the standard.

  NAL is responsible for the use of underlying network segment format to encapsulate data, including framing, logical channel signaling, timing information of use or end of the sequence signals, etc. NAL support, for example, video transmission format on circuit switching channels, support the video on the Internet use of RTP/UDP/IP transmission format. NAL includes its own header information, section structure and the actual load information, namely the upper VCL data. (if the data segmentation technology, the data may be composed of several parts).

  2,High precision, multiple patterns of motion estimation

  H. 264 support 1/4 or 1/8 pixel accuracy motion vector. In 1/4 pixel accuracy when using 6 tap filter to reduce the high frequency noise, the accuracy of 1/8 pixel motion vector, can use more complicated 8 tap filter. In the motion estimation, the encoder can also choose "enhanced" interpolation filter to improve prediction effect.

  Motion prediction in h. 264, a macro block (MB) in figure 2 is divided into different blocks, block size seven different model). This model more flexible and meticulous division, the shape of the moving object in the image more, greatly improving the accuracy of motion estimation. In this way, each macro block can contain 1, 2, 4, 8 or 16 motion vector.

  In h. 264, allows the encoder to use more than one frame of the previous frame is used for motion estimation, this is what is called a multiple reference frame technology. For example 2 frames or 3 frame just coding good reference frame, the encoder will choose macro block for each object can give better prediction of the frame, and instructions for each macro block is which a frame is used to predict.

  3, 4 * 4 block integer transform

  H. 264 is similar to the previous standard, the residual error based on block transform coding, but transform is integer operations instead of real number operation, the process is similar and DCT. The advantage of this approach is that: in the encoder and decoder is allowed in the accuracy of the same transform and inverse transform, simple easy to use fixed point operation mode. That is to say, there is no "inverse transform error". Transformation is the unit of 4 * 4 blocks, instead of commonly used 8 x 8 pieces. Reduces the size of block due to transform, the division of moving object is more accurate, in this way, not only transform the relatively small amount of calculation, and on the edge of the moving object in cohesive error is greatly reduced. In order to make the small size blocks in the way of transformation of image of large-area smooth area do not produce the grayscale difference between, can macroblock brightness data within the frame of 16 4 * 4 block DC coefficient (each small piece of a, a total of 16) on the second piece of the transformation of 4 * 4, 4 of chromaticity data of 4 * 4 block DC coefficient (each small piece of a, a total of 4) 2 x 2 pieces of transform.

  H. 264 in order to improve the ability of rate control, quantitative step change amplitude control at about 12.5%, rather than at a constant rate changes. The amplitude of the transform coefficient in the process of normalization is placed on the quantitative processing in order to reduce the complexity of calculation. In order to emphasize the colour lifelike, the chroma coefficient adopted smaller quantization step length.

  4, Unified VLC

  There are two kinds of entropy coding method in h. 264, is a symbol of all stay encoding USES the unified VLC (UVLC: UniversalVLC), another is to use content adaptive binary arithmetic coding (CABAC: Context - AdaptiveBinaryArithmeticCoding). CABAC is optional, and its coding performance than UVLC slightly better, but also has a high computational complexity. Use an infinite length code charset UVLC, design rules, is a lot of in the same clock can code for different objects. This method is easy to generate a code word, and decoder is easily prefix identification code word, UVLC when bit error can quickly get the synchronization.

  5, Frame prediction

  In the previous h. 26 x series and MPEG - x series standard, are all adopt the interframe prediction. In h. 264, when the Intra coding image frame could be used to predict. For each of the 4 * 4 blocks (in addition to the outside edge of special disposal), every pixel can be 17 closest previously encoded pixels of different weighted sum (some weights can be 0) to predict, namely the pixels in the top left corner of the piece of 17 pixels. Obviously, this kind of frame prediction is not on time, but the predictive coding algorithm in spatial domain, you can remove the spatial redundancy between adjacent blocks, get a more effective compression.

  As shown in figure 4, 4 * 4 block a, b,... , p 16 for the prediction of pixels, and A, B,... , P is encoded pixels. Such as the value of m point can be (J + 2 + 2 K + L) / 4 type to predict, can also by (A + B + C + D + I + J + K + L) / 8 to predict, and so on. Are chosen according to the forecast of the reference point is different, brightness in 9 different types of patterns, but only 1 of chromaticity frame prediction models.

  6,For IP and wireless environment

  H. 264 draft contains a tool for error elimination, convenient for compressed video in the error transmission, packet loss, multiple environments, such as mobile channel or IP transmission robustness in the channel.

  In order to resist the transmission error, h. 264 video stream of the time synchronization can be finished by using frame image refresh, space synchronization by article structure coding (slicestructuredcoding) to support. At the same time, to facilitate the error after resynchronization, in an image of the video data also provides a synchronization point. In addition, the frame macroblock refresh and multiple reference macroblock allows the encoder when decide the macroblock mode can not only consider the coding efficiency, also consider the characteristics of transmission channel.

  In addition to using the quantization step change to adapt to the channel rate, in h. 264, also often use data segmentation methods to cope with changes in the rate of channel code. Overall, the concept of data segmentation is generated in the encoder video data with different priorities to support quality of service QoS of network. Such as data segmentation based on syntax (syntax - baseddatapartitioning) method, each frame of data according to its importance can be divided into several parts, this allows the buffer overflow when discarded less important information. Also can be used in a similar time data segmentation (temporaldatapartitioning) method, through in the P frame and B frame using multiple reference frame.

  In the application of wireless communication, we can change the quantitative accuracy of each frame or space/time resolution in support of the wireless channel bit rate change. But, in the case of multicast, encoder to change different bit rates for response is not possible. Therefore, different from the fine classification coding in mpeg-4 FGS (FineGranularScalability) method (low) efficiency, h. 264 adopts flow switch SP frame instead of coding.


[S] [C] [B] [Close window]