Automatic classification of outdoor images by region matching

Abstract

This paper presents a novel method for image classification. It differs\nfrom previous approaches by computing image similarity based on region\nmatching. Firstly, the images to be classified are segmented into\nregions or partitioned into regular blocks. Next, low-level features\nare extracted from each segment or block, and the similarity between\ntwo images is computed as the cost of a pairwise matching of regions\naccording to their related features. Experiments are performed to\nverify that the proposed approach improves the quality of image classification.\nIn addition, unsupervised clustering results are presented to verify\nthe efficacy of this image similarity measure.