Incorporating efficiency and human judgment in image retrieval for trademark matching

Abstract

There are several studies in the literature comparing different approaches. Most of these comparisons are based on objective tests (i.e.; efficiency and effectiveness of the approaches are obtained and compared). In this paper we conducted a novel subjective test, where human perception is incorporated to the evaluation process. Five known methods in the image retrieval literature are implemented and compared for closeness to human perception and also for their search time. Here, 1) similarity, 2) symmetry, and 3) area of trademarks retrieved by five different methods are evaluated and scored by humans. Experimental results illustrate that the correlation method is the nearest to human's perception in all fields. Experiments also show that the EPNH method is more efficient (much more shorter time) than the correlation method, while the semantic powers of these two are close together.