Modeling of contours in wavelet domain for generalized lifting image compression

Abstract

This paper introduces the design of context-based models of contours in the wavelet domain, which are used to construct generalized lifting (GL) mappings for image compression. The GL context-based mapping may significantly reduce the signal energy and the resulting bitrate. Here, we propose a strategy to define a reduced set of structured models to design the GL. The models capture the contour structures and are contrast-invariant. Initial experimental results applying the strategy on a wavelet subband exhibit potential gains. Iterations of the GL scheme as well as an adaptive entropy coding strategy may increase the coding gain.

Generalized Lifting with adaptive local pdf estimation for image coding

“…Adaptive local pdf estimation approach to GL mapping construction has demonstrated to provide energy reductions comparable to those obtained through the use of models [8]; with the advantage of the independency of the DWT filter. As a counterpart, this approach is more sensitive to the wavelet domain subband structure.…”

“…As the number of context classes and of parameters used to define the conditional subband pdfs is small, this overhead has only a minor impact on the final bitrate. In [8], we show that this strategy provides significant gains in terms of energy reduction of the wavelet coefficients, in particular around contours.…”

“…This strategy is independent of the wavelet filters used to transform the image. Experimental results exhibit interesting gains in terms of energy reduction comparable to those obtained in [8]. In order to take benefit from this energy reduction, specific entropy encoder should be designed in the future.…”

“…We propose here an adaptive estimation strategy that essentially consists in searching in a causal neighborhood of the current sample to encode for contexts that are similar to the one being analyzed. The goal of this paper is to describe and analyze this approach and to compare its results with those reported in [8]. Note that the proposed approach eliminates the dependency of the GL mapping on the wavelet filters used in the initial DWT.…”

“…To be able to deal with arbitrary images, a modeling approach was proposed in [8]. The strategy consists in clustering the initially huge set of possible context values into a reduced number of context classes and in modeling the pdfs conditioned to the classes.…”