Applications of Multi-channel Filter Banks to Textured Image Segmentation

Abstract

Multi-channel filtering provides a robust solution to the texture discrimination/ segmentation problem for digital image processing. Specifically, the use of the Gabor filter for channel characterization is explored. Tag deformation analysis is used to construct a new tag tracking pre-filter based on the multi-channel paradigm. The new pre-filter is experimentally shown to be superior to two other currently used pre-filters. Also, the use of Gabor filters for channel characterization is applied to tagged cardiac MRI, cine cardiac MRI and segmentation of asphalt core images. In the applications involving MRI, segmentation algorithms are developed and tested for optimal parameters. The segmentation algorithm developed for asphalt core images is intended to be used in subsequent steps involving an aggregate orientation analysis.