Parallel multi-tree indexing for evaluating large descriptor sets

This paper presents a method towards easier evaluation of a large number of different image/video content descriptors, by using a multiple descriptor-tree based parallel indexing scheme instead of classical index structures with high dimensional multi-feature vectors. We will show that the proposed scheme is flexible and easily extensible, and it is not just faster to build, but provides good retrieval precision as well. The primary goal is to provide a flexible and modular indexing scheme for descriptor evaluation and feature selection purposes, but it can be used for generic content-based retrieval tasks as well.