Application of affine invariant Fourier descriptor to shape-based image retrieval

摘要

Contour-based descriptors are among the main approaches in content based image retrieval. Most of these descriptors are based on Fourier transform and use various shape signatures and retrieval methods. There retrieval rates are good, and they perform generally in speeded ways. This paper presents a new approach which looks able to upgrade these results. Prior to feature extraction, the shape undergoes moment-based preprocessing in order to ensure affine transformations robustness. A double signature is computed from shape radius and specific angles. Then, we compute the coefficients of Fourier descriptors, and with a specific similarity measure we get an efficient shape retrieval performance. Our approach is compared to a classical Fourier descriptor and to another variant using PCA. We also design a comparison with other shape contour-based descriptors.