Texture detection and texture discrimination with anisotropic scaling indices

A new method to determine specific local features of textures is proposed. Based on the idea of pointwise dimensions, the concept of isotropic and anisotropic scaling indices is developed. A new algorithm for texture detection and texture localization that makes use of level-a statistics is introduced. The performance of the algorithm is evaluated on an example of two natural textures. The effect of noise is studied, and it is shown that the algorithm is insensitive to white noise. The global and local aspects of the detection and discrimination process within this statistical model are pointed out. Some remarks on the choice of parameters in texture discrimination algorithms are made. (C) 1997 Optical Society of America.