Novel rate-distortion analysis framework for bit rate and picture quality control in DCT visual coding

The rate-distortion (R-D) behaviour of a coding system depends on both the characteristics of the input source data and the coding algorithm. The authors introduce the new concepts of characteristic rate curves and rate curve decomposition to characterise the input source data and model the coding algorithm, respectively. Based on these concepts, a novel framework for R-D analysis is developed, which serves as an alternative to the classical R-D analysis. Based on this framework, a fast algorithm is proposed to predict the R-D curve before quantisation and coding. The proposed algorithm has very low computational complexity. However, in the extensive simulations, its relative prediction error is always less than 5%, which is very small. To the best of the authors' knowledge, this is the first algorithm which is able to accurately estimate the R-D curve before quantisation and coding. With the estimated R-D curve, it is possible to accurately control the output bit rate and picture quality for transform coding of still images and video sequences. In practical visual coding applications, with the proposed R-D estimation algorithm, the coding bit rate can be accurately matched to the available network bandwidth to guarantee the successful transmission of the coded image/video data.