Quaternion representation of color and texture integration based image segmentation approaches

Abstract

Image segmentation is one of the most investigated subjects in the field of computer vision since it plays a crucial role in the development of high-level image analysis tasks such as object recognition and scene understanding. Segmentation of natural images is by far a more difficult task, since natural images exhibit significant inhomogeneities in color and texture. The use of color and texture information collectively has strong links with the human perception, but the main challenge is the combination of these fundamental image attributes in an integrated color-texture image descriptor. Therefore the present study aims at the development of simple, fast, efficient and fully automatic color texture segmentation algorithms. Three different new approaches are proposed in this study to address the problem of color texture segmentation. The important points that are considered in all the three approaches are: (i) More than one color space should be used and also quaternion representation of color can be used in one or other way. (ii) A criterion should be applied to determine the optimal number of clusters automatically. (iii) Both spatial and spectral information should be added to the segmentation process in order to reduce the effect of noise and intensity inhomogeneities introduced in imaging process. (iv)Adaptive inclusion of color and texture information, to favor the solution of piecewise - homogeneous labeling. Among the proposed approaches, the second approach namely Pixon representation and modified Gaussian mixture model based segmentation algorithm improved the PRI value to an extent of 6.5% and reduced the BDE error by 20% than the conventional approach CTM in the first experiment and also reduced the mean, standard deviation and r.m.s. errors than the conventional approaches in the second experiment. Hence, this proposed approach is found to be more dependable and accurate. newline newline newline