Scene Adaptive Parameters Selection for MPEG Syntax Based HDTV Coding

For a constant-bit-rate image coding algorithm, bit allocation at encoder plays a very important role in controlling the coding performance. In this paper, a two-pass MPEG syntax based video coding algorithm[1] with forward analyzer that takes into account the human visual system is discussed. The first pass is to estimate the bits and the quantization step sizes needed in coding based upon the desired visual quality and buffer status; the second pass is to really encode the picture with real time adjustment. A macroblock bit estimation unit that incorporates the human visual model is used to produce equal quality MB's in a frame, and a frame bit estimation unit that incorporates scene complexity is used to produce equal quality pictures in a video sequence. A scene analyzer is adopted to calculate the activity and the perceptual threshold of each MB and the scene complexity of an entire picture. Then, a frame bit allocation unit, which takes into account both the scene complexity and the buffer status, decides the bits needed in encoding this frame. Once the bits of a frame is determined, the MB bit allocation unit decides the quantizer scale for each MB subject to the estimated frame bits and the perceptual threshold. In the coding process, an buffer regulation unit further modifies the quantizer scale in order to compensate the deviation in estimation.