Texture Classification Based on Graph Diffusion Wavelet Scattering Transforms

Texture is a critical feature of images, and the graph scattering transform network provides a multi-scale analysis framework for texture image analysis by capturing multi-scale structural features through graph structure and maintaining stability against perturbations. In this paper, we propose a texture classification method based on graph scattering transform, utilizing a generalized diffusion matrix to construct a graph wavelet filter bank, and combining multiple graph representations to extract the image texture information. Moreover, subband pruning is performed during the process of the scattering transform. The proposed classification method is validated on three common texture datasets, and the experimental results demonstrate the effectiveness of our proposed method in achieving superior classification performance.