Distortion weighting based on biorthogonal wavelet gain in JPEG2000

Abstract

JPEG2000 is the latest standard in image compression with the Joint Photographic Experts Group (JPEG) committee ratifying Part 1, second edition of the standard in September 2004. The JPEG2000 compression system improves upon the original JPEG compression standard through the use of a wavelet transform and an Embedded Block Coding with Optimal Truncation (EBCOT) technique that allows for a compression advantage over JPEG of about 30 percent on average. EBCOT depends on an arithmetic entropy coder called the MQ Coder that processes data and outputs the final image bit-stream that is stored in a JPEG2000 file. After the MQ Coder, the bit-stream can be truncated at points defined during the encoding process, and this truncation may be done optimally to minimize the amount of distortion introduced to the decoded image. This optimal truncation relies on a distortion calculation for every bit that makes up an image. In order to properly calculate this bit-distortion, weights are applied to a squared-error calculation, scaling the resulting distortion in a manner that accounts for the gain associated with the biorthogonal wavelet transform used in JPEG2000. This paper gives a background on the derivation of these weights and where they are applied in the optimal truncation process.