Automatic and manual tattoo localization
Joonsoo Kim, Li He, Jiaju Yue, Javier Ribera, Edward J. Delp, Landis M. Huffman · 2016
In this paper we introduce two different automatic tattoo localization methods. The first one is a center-surround feature localization method. It combines a center-surround filter with skin and edge features based on the observation that the skin area surrounding the tattoo is homogeneously smooth and skin-colored. The second method is a graph-cut tattoo localization method. It detects the tattoo region using graph-cut segmentation based on image edges, a skin color model and a visual saliency map. In both methods tattoo regions are correctly detected even when a tattoo image includes background clutter. We also describe a manual tattoo localization tool that supports several functions to enable a trained annotator to crop and save tattoo regions efficiently. Experimental results show that our automatic methods achieve good performance on two different tattoo image datasets that include background clutter.