Spatio-spectral networks for color-texture analysis

Texture is one of the most-studied visual attributes for image characterization since the 1960s. However, most hand-crafted descriptors are monochromatic, focusing on grayscale images and discarding the color information. Therefore this work proposes a new method for color texture analysis considering all color channels in a more thorough approach. It consists of modeling color images as directed complex networks that we named Spatio-Spectral Network (SSN). Its topology includes within-channel connections that cover spatial patterns of individual color channels, while between-channel connections tackle spectral properties of channel pairs in an opponent fashion. Image descriptors are obtained through topological characterization of the modeled network in a multiscale approach with radially symmetric neighboring. Experiments with four datasets cover several aspects of color-texture analysis, and results demonstrate that SSN overcomes all the compared literature methods, including known deep convolutional networks. It also has the most stable performance between datasets, achieving 98.5(± 1.1) of average accuracy against 97.1(± 1.3) of MCND and 96.8(± 3.2) of AlexNet. Additionally, an experiment verifies the performance of the methods under different color spaces, showing that SSN presents the highest performance and robustness.