Analysis directional features in images using Gabor filters

Abstract

A computational technique for directional analysis of piece-wise linear patterns in images has been developed using Gabor filters. A given image is preprocessed by a sequence of Gabor filters, which are Gaussian-modulated sinusoids in the spatial domain. Linear segments at specific orientations are detected by Gabor filters tuned to the corresponding directions. Information regarding orientation of the linear patterns in the image and the area covered by the patterns in specific directions is then computed from the filtered images. The performance of the method is illustrated through applications to synthetic images and scanning electron microscope images of collagen fibrils in rabbit ligaments. The method has significant applications in quantitative analysis of ligament healing and comparison of treatment methods for ligament injuries.<>