Sparsity based segmentation in hybrid color space

Recently in signal processing, data models based on sparsity prior have drawn much attention. Using this prior several state-of-the-art result is produced in the case of image and video processing based applications. Furthermore, learning the model parameters greatly improves the performance of a given application. We have studied the learning of such models in relevant feature space, and applied them for color image texture segmentation. We have proposed a scheme for construction of feature vectors for dictionary learning in a sparse framework that enhances the performance of color segmentation. Experimental results validate the scheme adopted, in terms of segmentation efficiency.