Perceptually optimized JPEG2000 coder based on CIEDE2000 color difference equation

Abstract

In this paper, a perceptually optimized JPEG2000 coder for color image compression is proposed. Based on the CIEDE2000 color difference equation, this paper presents a visual model for estimating the perceptual redundancy inherent in each pixel of a color image in terms of a triple of values, one for each color component, as a just noticeable distortion (JND). By considering the varying sensitivities of the human visual perception to luminance and chrominance signals of different spatial frequencies, the full-band JND profile for each color channel is decomposed into component JND profiles for different wavelet subbands. With error visibility thresholds provided by the JND profile of each subband, the perceptually insignificant wavelet coefficients in three color channels are first removed. Without altering the format of the compressed bit stream of the JPEG2000, the encoder is modified in a way that the bit rate is inversely correlated with the perceptible distortion rather than the distortion of mean square errors. In each coding pass, each coding block is monitored to stop the coding process if the perceptible distortion defined by the corresponding JND is minimized for a given bit rate. Simulation results show that the performance of the perceptually tuned coder in terms of the bit rate required for the same subjective visual quality and the amount of perceptual distortion at the same bit rate is superior to that of the un-tuned JPEG2000 coder.