Analysis of H.264 and AVS Core Technologies

H.264, like previous standards, still uses a hybrid coding framework. The AVS video standard uses a similar technical framework to H.264, including transformation, quantization, entropy coding, intra-frame prediction, inter-frame prediction, loop filtering and other modules. The differences in their core technologies include the following:

　　1. Transformation and Quantization

　　H.264 uses block-based transform coding for residual data to remove spatial redundancy of the original image, so that the image capacity is concentrated on a small part of the coefficients. The DC coefficient value is generally the largest, which can improve the compression ratio and enhance the anti-interference ability. The previous standard generally uses DCT transform. The disadvantage of this transform is that there will be a mismatch phenomenon. There will be a difference after the original data is transformed and restored by inverse transform. Since it is a real number operation, the amount of calculation is also relatively large. H.264 uses integer transform based on 4×4 blocks.

　　AVS uses 8×8 integer transform, which can be implemented on a 16-bit processor without mismatch. It is more effective than 4×4 transform in decorrelation of high-resolution video images, and uses 64-level quantization to meet the requirements of different applications and services for bit rate and quality.

　　2. Intra-frame prediction

　　Both H.264 and AVS technologies use intra-frame prediction, using adjacent pixels to predict the current block, and using multiple prediction modes that represent spatial domain textures. H.264 has two prediction modes for luminance prediction: 4×4 blocks and 16×16 blocks. For 4×4 blocks: there are 9 prediction directions from -135 degrees to +22.5 degrees plus a DC prediction; for 16×16 blocks: there are 4 prediction directions. Chroma prediction is 8×8 blocks, with 4 prediction modes, similar to the 4 modes of intra-frame 16×16 prediction, where DC is mode 0, horizontal is mode 1, vertical is mode 2, and plane is mode 3.

　　3. Inter-frame prediction

　　H.264 inter-frame prediction is a prediction mode based on coded video frames and block-based motion compensation. The difference from previous standard inter-frame prediction is the wider range of block sizes, the use of sub-pixel motion vectors and the use of multiple reference frames.

　　H.264 has 8 types of macroblock and sub-macroblock divisions: 16×16, 16×8, 8×16, 8×8, 8×4, 4×8 and 4×4, while AVS has only 4 types of macroblock divisions: 16×16, 16×8, 8×16 and 8×8.

　　H.264 supports the use of multiple different reference frames to predict inter-frame macroblocks and slices. In AVS, P frames can use up to two forward reference frames, and B frames use one reference frame before and after.

　　4. Entropy Coding

　　H.264 has formulated entropy coding efficiency based on the amount of information. One method is to use a unified variable length coding (UVLC) for all symbols to be encoded, and the other is to use context-adaptive binary arithmetic coding (CABAC) based on content, which greatly reduces the block coding correlation redundancy and improves the coding efficiency. UVLC has a low computational complexity and is mainly used for applications with strict coding time. Its disadvantage is low efficiency and high bit rate. CABAC is a very efficient entropy coding method, and its coding efficiency is 50% higher than that of UVLC coding.

　　AVS entropy coding uses adaptive variable length coding technology. In the AVS entropy coding process, all syntax elements and residual data are mapped into binary bit streams in the form of exponential Golomb codes.

　　The advantages of using exponential Golomb codes are: on the one hand, its hardware complexity is relatively low, and the code can be parsed according to a closed formula without looking up a table; on the other hand, it can flexibly determine the K-order exponential Golomb code encoding based on the probability distribution of the coding elements. If K is selected appropriately, the coding efficiency can approach the information entropy.

　　The block transform coefficients of the prediction residual are scanned to form (level, run) pairs. Level and run are not independent events, but are strongly correlated. In AVS, level and run are 2D jointly encoded, and the order of the exponential Golomb code is adaptively changed according to the different probability distribution trends of the current level and run.

　　In addition, there are no SI and SP frames in AVS. It can be said that AVS was developed on the basis of H.264 and absorbed the essence of H.264, but in order to bypass the patent problem, it had to abandon some core algorithms of H.264. The price in return is that the complexity is greatly reduced while the coding efficiency is slightly reduced.