Face sketch recognition via data-driven synthesis

Abstract

In some real-world scenarios, there does not always exist a normal photo for face recognition or retrieval purpose, e.g. suspect searching for law enforcement. Under the circumstances, a sketch drawn by the artist is usually taken as the substitute for matching with the mug shot photos collected by the police office. However, due to the great discrepancy of the texture presentation between sketches and photos, common face recognitionmethods achieve limited performance on this task. In order to shrink this gap, sketches can be transformed to photos relying on some machine learning techniques and then synthesized photos are utilized to match with mug shot photos. Alternatively, photos can also be transformed to sketches and the probe sketch drawn by the artist is matched with the transformed sketches subsequently. Existing learning-based face sketch–photo synthesis methods are grouped into two major categories: data-driven methods (example-based methods) and model-based methods. This chapter would give a comprehensive analysis and comparison to advances on this topic.