Shoeprint Image Retrieval by Topological and Pattern Spectra

In this paper, we propose a novel technique for the automatic classification of noisy and incomplete shoeprint images, based on topological and pattern spectra. We first consider the pattern spectrum proposed by Maragos. We extend each spectrum with the spectrum for the complement image. We also propose a topological spectrum for a shoeprint image, based on repeated open operations with increasing size of structuring element, giving a distribution of Euler numbers. The normalised differential of this series gives the topological spectrum. We secondly propose a hybrid algorithm which uses a distance measure based on a combination of both spectra as the feature of a shoeprint image. To evaluate the performance of the techniques, we use a database of 500 'clean' shoeprints to generate five test databases each with 2500 degraded images, such as Gaussian noise, incompletion, rotation, rescale, and scene background. The statistical evaluations in terms of precision vs. recall are given in the final section. Tests show that our hybrid technique combining both spectra gives significant improvements over previously published results for edge direction histogram.