A scale invariant distance measure for texture retrieval

We propose a similarity measure between two textures based on moments of the Fourier magnitude spectrum. The resulting distance is robust to changes in scale as well as to spatial shifts and grey-scale transforms of the texture samples. This type of invariant distance has applications to content-based image retrieval and classification tasks. We test the performance of the algorithm in a retrieval scenario using texture patches from the Brodatz album. The results indicate that the distance measure emulates human similarity perception in comparing textures.