Retrieval of hand-sketched envelopes in logo images

Abstract

This paper introduces an approach for retrieving envelope of high-level object groupings in bi-level images with multiple objects. Motivated by studies in Gestalt theory, hierarchical clustering is used to detect the envelope and group its objects based on their spatial proximity, area, shape features and orientation. To decide the final grouping, the grouping outcomes are combined using an evidence accumulation method. The high-level boundary of the detected envelope is then extracted using morphological operations. For retrieval, the boundary of a query sketch is matched to the extracted envelopes from database images via dynamic space warping. Experiments on a set of bi-level logo images demonstrate the effectiveness of the approach. © Springer-Verlag Berlin Heidelberg 2007.