Texture features are among the most commonly used image attributes in image understanding applications, such as image retrieval from databases. A number of methods and their variants have been developed over the years for texture feature extraction. Whereas they all have their merits and flaws, it is worthwhile to evaluate their performance in a specific application domain. The goal here is to establish which texture features are better suited for segmentation of natural scenes that contain multiple natural and synthetic textures. The performance of unsupervised texture segmentation based on multiresolution simultaneous autoregressive (MRSAR) models, wavelet coefficients, fractal dimension, edge direction and magnitude, and color moments is examined.
Department, Program, or Center
Chester F. Carlson Center for Imaging Science (COS)
IEEE Western New York Image Processing Workshop 2001
RIT – Main Campus