A color spatio-temporal segmentation tool applied to sequences of color texture images

In this paper, we present a new approach to discriminate dynamic color textures in sequences of images containing moving objects. The moving textures are analyzed by use of Haralick features extracted from color spatio-temporal co-occurrence matrices which characterize the textures themselves as well as their movements. Features with the highest discriminating power are selected according to a supervised learning scheme that permits to represent the dynamic color textures in a relatively small dimensional feature subspace where they can be easily discriminated. After testing and validating this approach on image sequences generated from the VisTex database, it has been applied on sequences of submarine images in order to automatically identify the red alga.