Combining Unsupervised Clustering with a Non-Linear Deformation Model for Efficient Petroglyph Recognition

Petroglyphs are prehistoric engravings in stone unrevealing stories of ancient life and describing a conception of the world transmitted till today. In the current paper we consider the problem of developing tools that automate their recognition. This is a challenging problem mainly due to the high level of distortion and variability of petroglyph reliefs. To address these issues, we pro- pose a two-stage approach that combines unsupervised clustering, for quickly obtaining a raw classification of the query image, and a non-linear deformation model, for accurately evaluating the shape similarity between the query and the images of the more appropriate classes.