Rate-distortion based scalable progressive image coding

The emergence of distributed, heterogeneous media such as the Internet has established the practical importance of progressive image transmission, in which an image is transmitted in such a way as to admit coarse rendering and recognition at the decoder as early as possible in the bitstream. This paper presents a wavelet-based progressive image transmission algorithm that attempts to achieve several goals not addressed by other image compression algorithms in the literature. First, the algorithm evaluates the tradeoff between rate and distortion as a criterion for selecting wavelet coefficients for transmission. The distortion metric is not limited to mean squared error; the algorithm provides a framework for investigating any distortion function including psychovisual and segmentation-based distortion metrics. Second, it provides a high degree of spatial scalability by sending coarser resolution information earlier in the bitstream than detail information and does not waste bits by refining high frequency subbands early. Finally, the algorithm is computationally asymmetric, pairing a very fast decoder with an encoder that can be as computationally intensive as required. The performance of the algorithm is comparable with current coders at low bitrates.