Multiscale Image Fusion Using the Undecimated Wavelet Transform With Non-Orthogonal Filter Banks

In this work a pixel-level multiscale image fusion framework is introduced which utilizes a new class of non- orthogonal filter banks. These filter banks were especially de- signed for the purpose of image fusion and offer some useful properties such as a short support size of the analysis filter pair whilst preserving the regularity of the synthesis filters. As a result, the unwanted spreading of coefficient values around overlapping image singularities, usually complicating the feature selection process, is minimized. Furthermore, the introduction of blocking artifacts in the fused reconstruction is avoided. In the course of this work we will show that the combination of the Undecimated Wavelet Transform with such filter banks leads to a fusion framework which is able to significantly outperform traditional multiscale fusion approaches for a large group of images, derived from different sensor modalities.