A Transfer Learning Approach for the Tattoo Classification Problem

Object classification is a widely studied topic in computational vision, with several areas of application. However, when it comes to images that contain tattoos, current models do not fit properly and, in general, they cannot correctly classify the tattoo elements. Thus, we present a deep learning model based on transfer learning for the tattoo classification problem. Furthermore, given that datasets in this area are very rare, a new dataset with 40 tattoo classes was created and presented for the proposed study. We also used data augmentation to improve the diversity of the training sets to achieve better classification accuracy. As a result, the proposed model demonstrated robustness in classifying tattoo images with an accuracy of 85.24% and, even in tests with images of unknown classes, the model proved to be comprehensive enough to classify these images as the semantically similar classes known by the model.