Efficient Image Clustering using a New Image Distance

A new distance for image clustering called Generalized Geodesic Distance (GGD) and an appearance-based image clustering approach called Global Geometric Clustering for Image (GGCI) are presented. Unlike the traditional distance, GGD takes into account the spatial relationships of images. Therefore, it is robust to small perturbation of images. GGCI based on GGD uses easily measured local metric information to learn the underlying global geometry of images space, then applies the extended nearest neighbor approach to cluster images. Different from the usual nearest neighbor approach, GGCI considers the density around the nearest points within manifolds embedded in high dimensional image space, which better reflects the intrinsic geometric structure of manifold. Experimental results suggest that the proposed GGCI approach achieves lower error rates in image clustering.