Simulation of text tattoo for text tattoo detection

Abstract

Tattoo is considered as a soft biometric trait and tattoo studies usually focus on tattoo detection and recognition for forensic applications. However, limited number of real text tattoo images may affect the accuracy of tattoo detection and recognition. Hence, in this project, we first generate pure text images and then experiment with simulation of text tattoo images for text tattoo detection. It takes a pure text image, an input skin image with its skin segmentation result, a manual selection of a large quadrilateral from the skin area in the input skin image, and some parameters as input. Then we apply digital tattooing based on the digital tattooing algorithm for portrait tattoo proposed in a related work. During the process, we randomly generate a skin mask based on the manual selection of the large quadrilateral from the skin area. The colour of the tattoo text is randomly selected from a colour pool obtained from the input skin image. Different parameters are also adjusted to ensure visibility of tattoo text and make the simulated text tattoo more realistic compared with the real text tattoo images.