A Detection-Theoretic Approach to Texture and Edge Discrimination

We present a probabilistic approach to the discrimination between textured areas and edges; locally defined probabilistic models are used, which model textured areas as sinusoidal and edges as phase-congruent signals. We build a link with energy-based feature detection and propose a simple approach to the discrimination between these two classes. This facilitates the selective use of texture/edge features for edge detection and image segmentation applications. Our experimental results demonstrate the applicability of our method on natural images.