Automatic human hallmark recognition based on visual words

Abstract

Human hallmarks, as an important biometric traits, contain discriminative characteristics information to support person identification due to its saliency in visual attention. A sketch of a tattoo can be drawn based on the description provided by an eyewitness or the victim. In this paper, the authors propose a novel hallmark photo-sketch recognition method using a bag-of-words (BoW) methodology for retrieving the photos in the database based on a query sketch or image drawn by an artist. And this article discusses the number of visual words generated which is based on the number of clusters, including vocabulary size, weighting scheme. Extensive experiments are conducted on a sketch database including 100 tattoo sketches drawn by three different subjects. Each sketch has a corresponding tattoo image, which is downloaded from the internet. It is an effective and viable application to develop operationally-relevant automated image-based tattoo recognition applications for pertinent law enforcement operational scenarios.