How to make local image features more efficient and distinctive
Chunxia Duan, Xiangfeng Meng, Changhe Tu, C. Yang · IET Computer Vision · 2008
A technique to construct efficient and distinctive descriptors for local image features is presented. The authors start with the scale invariant features detected and the gradient data of their neighbourhood patches in suitable size normalised and then apply independent component analysis (ICA) to obtain the independent components of the feature patches. The authors show how the ICA technique could be used to encode the salient aspects of the feature vectors because of the high-order statistical characteristics of both natural images and ICA. Comparisons are made between our descriptors and some state-of-the-art methods (e.g. scale invariant feature transform). Experimental results demonstrate that the proposed local feature descriptor is distinctive and with high matching speed.