Fast and accurate texture-based image segmentation

In this paper, the development and application of a fast algorithm for segmentation of textured images is discussed. It is based on Markov random fields as a method of feature extraction. We present a post-processing algorithm which increases the classification accuracy of an initial pixel-by-pixel scheme. The algorithm employs a majority decision concept to counteract the misclassification caused by multiple textures in a computational window. The method is then extended to yield a high speed algorithm which combines pixel and region classification, affording large computational savings. Experiments for both synthetic and real images, yielding accurate results, are reported