Skipping Selected Steps of DWT Computation in Lossless JPEG 2000 for Improved Bitrates.

Roman Starosolski,Li Zeng

doi:10.1371/journal.pone.0168704

Abstract

In order to improve bitrates of lossless JPEG 2000, we propose to modify the discrete wavelet transform (DWT) by skipping selected steps of its computation. We employ a heuristic to construct the skipped steps DWT (SS-DWT) in an image-adaptive way and define fixed SS-DWT variants. For a large and diverse set of images, we find that SS-DWT significantly improves bitrates of non-photographic images. From a practical standpoint, the most interesting results are obtained by applying entropy estimation of coding effects for selecting among the fixed SS-DWT variants. This way we get the compression scheme that, as opposed to the general SS-DWT case, is compliant with the JPEG 2000 part 2 standard. It provides average bitrate improvement of roughly 5% for the entire test-set, whereas the overall compression time becomes only 3% greater than that of the unmodified JPEG 2000. Bitrates of photographic and non-photographic images are improved by roughly 0.5% and 14%, respectively. At a significantly increased cost of exploiting a heuristic, selecting the steps to be skipped based on the actual bitrate instead of an estimated one, and by applying reversible denoising and lifting steps to SS-DWT, we have attained greater bitrate improvements of up to about 17.5% for non-photographic images.

Highlights

In lossless JPEG 2000, the 5×3 kernel reversible discrete wavelet transform (DWT) performed using lifting steps decomposes an image into subbands of different characteristics, that are independently entropy coded [1, 2]
We found that the noise filtering was the most effective in improving lossless JPEG 2000 bitrates when applied during computing of some reversible denoising and lifting steps (RDLS)-modified DWT (RDLS-DWT) subbands only
That in the RDLS-modified DWT (RDLS-DWT) the noise filtering significantly improved the lossless JPEG 2000 bitrates of non-photographic images and of images contaminated with impulse noise; filtering was the most effective when applied to some subbands only

Summary

Introduction

In lossless JPEG 2000, the 5×3 kernel reversible discrete wavelet transform (DWT) performed using lifting steps decomposes an image into subbands of different characteristics, that are independently entropy coded [1, 2]. We found that the noise filtering was the most effective in improving lossless JPEG 2000 bitrates when applied during computing of some RDLS-modified DWT (RDLS-DWT) subbands only. Since in some cases the best bitrates were obtained when the DWT stage of JPEG 2000 was skipped, we suspected that to denoising, the optimum might be in-between skipping and applying DWT. In this study we propose the skipped-steps DWT (SS-DWT), which is obtained from DWT by skipping selected steps of its computation. By selecting steps to be skipped we may obtain, as special cases of SS-DWT, an unmodified DWT (if we do not skip any step), skip the entire DWT (by skipping all the steps of the transform), or skip the DWT partially

Objectives

Methods

Findings

Discussion

Conclusion