Performance Comparison of Wavelet Families for DWT-Based Fusion of CT & MRI Images

EasyChair Preprint 6933

Abstract

Research about image fusion has grown in the last three decades, essentially because there's a need to improve the data used for computerized post-processing and human visualization, being medical imaging a natural application for it. One of the latest used algorithms is the discrete wavelet transform (DWT) that is used for the analysis of non-stationary signals or those with discontinuities. Nevertheless, there are several wavelets that can be used as bases over which to do the transform. The preference of one over the other is determined experimentally and is heavily dependent on the application; in this case, the fusion of images obtained by magnetic resonance and computerized tomography. The fusion process with the 2D-DWT was done on MATLAB 9.3 CLI, using three pairs of slices for one, two, three and four levels of decomposition. The tested wavelets are symlet5, daubechies5 and discrete Meyer. The symlet5 filters showed the best (higher) mutual information (with superiority of ~0.07%) for all slices and almost all levels of decomposition except for level 2, in which discrete meyer wavelets were superior by 0.17%. For every pair of slices, symlet5 showed the best (lower) joint entropy. The results can be ascertained for T2 MRI and CT images.

Keyphrases: Entropy, Wavelet, mutual information

@booklet{EasyChair:6933,
  author    = {Diana Madrid},
  title     = {Performance Comparison of Wavelet Families for DWT-Based Fusion of CT & MRI Images},
  howpublished = {EasyChair Preprint 6933},
  year      = {EasyChair, 2021}}