Dirichlet Series in Complex Network Modeling of Texture Images

Abstract

This work investigates the use of Dirichlet series in the modeling of texture images, with application in image classification. The proposed model is based on a strategy that associates each pixel with its corresponding color (gray level in our case) to a vertex of a complex network and the gray level dissimilarity within neighbor pixels with edge weights. The degree distribution of such network is known to be very effective in providing image descriptors. Here, we propose an improvement over this technique, by working on this distribution as a Dirichlet (exponential) series and varying the exponential parameter. A family of statistical measures are extracted from the series and compose a feature vector employed here for texture image classification. In our tests, the achieved accuracy is promising when compared with other state-of-the-art approaches in different databases classically used for benchmark purposes.