A string matching approach for visual retrieval and classification

We present an approach to measuring similarities between visual data based on approximate string matching. In this approach, an image is represented by an ordered list of feature descriptors. We show the extraction of local features sequences from two types of 2-D signals { scene and shape images. The similarity of these two images is then measured by 1) solving a correspondence problem between two ordered sets of features and 2) calculating similarities between matched features and dissimilarities between unmatched features. Our experimental study shows that such a globally ordered and locally unordered representation is more discriminative than a bag-of-features representation and the sim- ilarity measure based on string matching is efiective. We illustrate the application of the proposed approach to scene classification and shape retrieval, and demonstrate superior performance to existing solutions.