Texture segmentation algorithm using multichannel wavelet frame

Abstract

Describes a segmentation algorithm for textured images by using a multichannel wavelet frame. An overcomplete wavelet frame transform is adopted to decompose a textured image into multichannel images. The method for extracting texture features is based upon an adaptive noise smoothing concept which considers the nonstationary nature of the noise filtering problem. Furthermore, this method incorporates contextual/spatial information among feature images to reduce variability of texture feature estimates while retaining the accuracy of region borders. In our segmentation system, the estimated feature vector of each pixel is sent into a Bayes classifier to make an initial probabilistic labeling. Then, the spatial constraints are enforced through the use of a probabilistic relaxation algorithm. Finally, the performance of the proposed segmentation system for textured images is demonstrated experimentally and comparisons of performance are made.