Textured Image Retrieval by the Modal Analysis of Quantized Spectral Point Patterns

In this work we explore the modal correspondence method of Shapiro and Brady to match points set by comparing the eigenvectors of a pairwise point proximity matrix for peak cluster centers in the power spectrum. We use a variant of the Carcassoni, Ribeiro and Handcok method to perform recognition. The new technique developed is invariant to rotational, scaling transformations, and is robust in the presence of noise.