Multi-channel Gabor Wavelet Filter-based Image Segmentation

Inspired by the multi-channel operation of the Human Visual System for interpreting texture[10], research has been focused on a multi-channel approach based on Gabor filtering to mimic the operation of HVS(Human Visual System)[1]. In this paper we employ this multi-channel approach to image segmentation in order to gain insight into the ability of this methodology in solving segmentation problem.The status of multi-channel approach based on Gabor filtering applied in image segmentation field recently is presented, and the theories, steps, results and analyses of several Gabor filters applied in image segmentation are given.Algorithms and analyses about edge detection and threshold selection of image segmentation are presented. An overview of the basic theories and the recent development is given, and two new multi-channel approaches based on Gabor filtering applied in image segmentation are presented.Gabor filters have the ability to perform multi-resolution decomposition due to its localization both in spatial and spatial frequency domain. Normally the effective width of a filter in the spatial domain and its bandwidth in the spatial-frequency domain are inversely related according the 2D uncertainty principle. That is why Gabor filters are well suited for image segmentation problem. In this paper we design and implement a image segmentation program based on the work done by Yossi Rubner[2].We employ two new approaches to the image after preprocessing the original image properly. One is the linear combination of the features extracting from the image using the Gabor filters. The other is using Gabor filters together with Sobel or Canny. We have had computer simulating experiments which prove that the multi-channel approaches to image segmentation achieves a good result and make preparations for the follow up.Programs were all compiled in the Win2000 by Matlab6.1. The programs can change scales and orientations of the Gabor filters and extract the features of the image and so on. All pictures were collected from Internet.