Intelligent rate control for MPEG-4 coders

The multimedia technology, which will lead in the 21st century, will depend on how to manipulate visual information efficiently, and will be focused on the interactive visual information exchange and user interactions. Among one of those technologies, the MPEG-4 codec can multiplex a set of independently-coded, arbitrarily-shaped video objects and transmit through either fixed or variable rate channels such as internet, wireless or satellite communications. Some quality control algorithms should support the encoding of visual objects to obtain and maintain best picture quality under the constraints of the quality requirements and channel environments. Especially, the MPEG-4 codec should be robust with respect to rapid changes in size and shape of the objects. The paper focuses on the design of the intelligent rate control algorithms via introducing quadratic neural networks and evaluating data-driven pattern analysis rather than rate-distortion mathematical models. According to data-driven pattern analysis, it is found that several new variables such as motion vectors are required to control near-optimally for transmitting best quality of moving pictures in real-time. We also found that the simplified mathematical rate distortion models, which are now widely used, could not support the control mechanism enough. Therefore, a quadratic neural-net using density estimation and randomizing pattern space, called density-based random-vector functional-link net is introduced to control the picture quality optimally. The proposed algorithm is tuned to obtain near-optimal picture quality of QCIF (176X144 pixels) format video streams for low bitrate transmission, storage/retrieval. Experimental work is presented based on the MPEG-4 video coder specification, apart from making intelligent rate control algorithm. The comparisons between the results of intelligent control and conventional rate control are presented.