1, the hierarchical design of the H.264 ip camera algorithm can be conceptually divided into two layers: video coding layer (VCL: Video Coding Layer) is responsible for efficient video content, said the network abstraction layer (NAL: Network Abstraction Layer) is responsible for network requirements of the appropriate way to package and transfer data. Between the VCL and NAL defines the interface a packet-based methods, packaging, and NAL corresponding signaling is part of. Thus, high coding efficiency and network friendly task by VCL and NAL to complete. VCL layer including block-based motion compensation hybrid coding, and some new features. And, like the previous video coding standard, H.264 is not the pre-treatment and post-processing functions included in the draft, which can increase the flexibility of the standard. NAL is responsible for the network segment using the underlying format to encapsulate data, including framing, logical channel signaling, timing or sequence of the end use of the information signals. For example, NAL supports circuit-switched video channel in the transmission format, support video on the Internet using RTP / UDP / IP transmission format. NAL header information, including their own, segment structure information and the actual load information, that the top of the VCL data. (If you use data partitioning, data may consist of several components).
2, high-precision, multi-mode motion estimation
H.264 support 1 / 4 or 1 / 8 pixel precision motion vector. 1 / 4 pixel precision 6-tap filter can be used to reduce high frequency noise, for 1 / 8 pixel precision motion vector can be used more complex 8-tap filter. During motion estimation, the encoder can also choose to "enhance" interpolation filter to improve the prediction results. Motion estimation in H.264, a macroblock (MB) 2 can be divided into different sub-block, the formation of seven different modes of block size. This multi-mode flexible and detailed division of the image more in line with the actual shape of moving objects, greatly improving the accuracy of motion estimation. In this way, in each macro block can contain a 1,2,4,8 or 16 motion vectors. In H.264, which allows the encoder to use more than one previous frame for motion estimation, which is called multi-frame reference technology. For example, two or three just a good reference frame encoding, the encoder will select for each target macroblock can give better prediction frame, and instructions for each macro block is a frame which is used for prediction.
3, 4 × 4 integer transform blocks
H.264 is similar to the previous standard, based on the residual block of transform coding, but the transformation is an integer operations rather than real operations, similar to the process and DCT. The advantages of this approach: in the encoder and decoder to allow the same precision transform and inverse transform, easy to use a simple fixed-point operation mode. In other words, there is no "anti-conversion error." Transformation unit is 4 × 4 blocks, rather than in the past used the 8 × 8 block. As the block size used to transform the narrow, more precise classification of moving objects, so that not only transform the calculation than the smaller, and the edges of moving objects in the interface errors are greatly reduced. In order to transform the way small pieces of a larger area of the image in the smooth area between the gray block does not produce differences in brightness can frame macroblock data of 16 blocks of 4 × 4 DC coefficients (each piece a total of 16) for the second 4 × 4 block transform, the color data of four 4 × 4 block of DC coefficients (one for each small piece of a total of four) for 2 × 2 block transformation. H.264 rate control in order to improve the ability to quantify the magnitude of step change in control of about 12.5%, rather than constant growth changes. Transform coefficients of normalized amplitude on the inverse quantization process is treated to reduce the computational complexity. To emphasize the color fidelity of the color factor with a smaller quantization step.