Texture Classification using Histogram of Sparse Representations

Texture classification is one of the major challenges in the field of texture analysis. It is a difficult task because of variations in the appearance of texture due to changes in scale, orientation, and exposure to light. Successful segmentation and classification of an image depends on being able to describe and classify the underlying texture efficiently. This document describes a method for the efficient description of texture in images. This method builds a dictionary of texture descriptors from a limited sample of images that can then be used to model textures in larger sets of images. The descriptors were obtained using gabor filters since it captures features in multiple scales and orientations. This enables the application of this method on images of different magnifications. This model is evaluated on a variety of natural images.