Affine invariant shape descriptors: The ICA-Fourier descriptor and the PCA-Fourier descriptor

In this paper, we propose two new affine invariant shape descriptors, the ICA-Fourier descriptor and the PCA-Fourier descriptor. We tested the descriptors by using them as features for shape based silhouette image retrieval. Experiments on a 1000 silhouette image database show promising retrieval rates of 95.41% and 93.63%, using the ICA-Fourier descriptor and the PCA-Fourier descriptor, respectively. The relationship between those two descriptors are also explained. The proposed PCA-Fourier descriptor is computationally more efficient than its ICA counterpart, while having comparable performance.