An Environment to Test Progressive Refinement of Indexing for Content-Based Image Retrieval

摘要

Content-based image retrieval is a fairly new discipline. Yet research in this field has highlighted many approaches that show good performance in specific sub-problems using single filtering criteria, such as color based histograms, shape description or texture analysis. Most systems built so far use multiple access methods although they compose them according to a pre-defined strategy. In this paper we report an environment that allows the evaluation of different type of filtering composition strategies. The aim is to let the developers work with a quick tool for fine tuning of sequences or compositions of methods, especially when such methods are based on different features. The selectivity of each indexing method is tested against a single database of images and then the overall performance of a sequence or a composition of methods is quickly estimated. The experiments have been applied on filtering criteria that share a multiresolution approach, in order to highlight the role that the hierarchical indexing approach can play in reducing the overall computational cost.