Application of reversible denoising and lifting steps to DWT in lossless JPEG 2000 for improved bitrates

Abstract

In a previous study, we noticed that the lifting step of a color space transform might increase the amount of noise that must be encoded during compression of an image. To alleviate this problem, we proposed the replacement of lifting steps with reversible denoising and lifting steps (RDLS), which are basically lifting steps integrated with denoising filters. We found the approach effective for some of the tested images. In this study, we apply RDLS to discrete wavelet transform (DWT) in JPEG 2000 lossless coding. We evaluate RDLS effects on bitrates using various denoising filters and a large number of diverse images. We employ a heuristic for image-adaptive RDLS filter selection; based on its empirical outcomes, we also propose a fixed filter selection variant. We find that RDLS significantly improves bitrates of non-photographic images and of images with impulse noise added, while bitrates of photographic images are improved by below 1% on average. Considering that the DWT stage may worsen bitrates of some images, we propose a couple of practical compression schemes based on JPEG 2000 and RDLS. For non-photographic images, we obtain an average bitrate improvement of about 12% for fixed filter selection and about 14% for image-adaptive selection.