Convolution implementation with a novel approach of DGHM multiwavelet image transform

Ondrej Kovac – Jan Mihalik – Iveta Gladisova

   The purpose of this paper is to develop convolution implementation of DGHM (Donovan, Geronimo, Harding, Massopust) multiwavelet image transform using a new approach of ordering wavelet coefficients at the second and higher levels. Firstly, the method of implementation of one-dimensional discrete multiwavelet transform (1D DMWT) for DGHM multiwavelet using discrete convolution and scalar filters is presented. Then, convolution implementation of DGHM multiwavelet image transform by application of 1D DMWT for two dimensions (2D) in a separable way is proposed. Next, the second level of 2D DMWT is performed in three possible ways. The novelty of the proposed implementation is in reordering of L subband wavelet coefficients at the first level into one subimage. The results are evaluated as the energy ratios between the transformed images in L subband at the second level and the input original image. According to the experimental results, the developed implementation of 2D DMWT is approximately 5% more effective in energy compression than the ones most commonly mentioned in the literature. This paper shows a possibility of convolution implementation of 2D DMWT with higher energy compression.

Keywords: multiwavelet transform, convolution, prefilter, postfilter, multifilter, compression

[full-paper]