Texture detection by genetic programming

This paper presents an approach to blind texture detection in images based on adaptation of the 2D-lookup algorithm by genetic programming. The task of blind texture detection is to separate textured regions of an image from non-textured (as e.g. homogeneous) ones, without any reference to a priori knowledge about image content. The 2D-lookup algorithm, which generalizes the well-known co-occurrence matrix approach of texture analysis, is based on two arbitrary image processing operations. By genetic programming, those image operations can be designed and adapted to a given recognition goal of the whole algorithm. The idea to employ such a framework for texture detection is to use a random image as adaptation goal. Despite of the fact that such a task has no exact solution, the system is able to fulfill this task to a certain degree. This degree is related to textureness in the image: the more texture, the higher the degree. The paper exemplifies this approach.