Efficient Image Feature Combination with Hierarchical Scheme for Content-Based Image Management System

This paper proposes efficient image feature combinations based on local descriptor and hierarchical indexing scheme obtained by clustering with global descriptor for content-based image management system such as image identification and identical image grouping. As features for the image retrieval, we consider both global feature which has general information of overall image for fast image retrieval and local feature which is based on feature points and has high matching accuracy for fine matching of images. The developed local feature is invariant to image scale and rotation, addition of noise, and change in illumination, thus, it sufficiently performs reliable matching between different views of scene across affine transformation. The method works with global feature among image clusters of database in advance and do fine searching only among image data in the cluster with local feature. In order to decrease computation time, we apply conventional clustering methods to group images similar in their characteristics together so that search can be made in a hierarchical manner by fine matching within partial database of candidate images. It can overcome the drawback of exhaustive matching time between similar images by using only local descriptor.