Face Sketch Recognition from Local Features

Abstract

Systems for face sketch recognition are very important for law enforcement agencies. These systems can help to locate or narrow down potential suspects. Recently, various methods were proposed to address this problem, but there is no clear comparison of their performance. In this paper is proposed a new approach for photo/sketch recognition based on the Local Feature-based Discriminant Analysis (LFDA) method. This new approach was tested and compared with its predecessors using three differents datasets and also adding an extra gallery of 10,000 photos to extend the gallery. Experiments using the CUFS and CUFSF databases show that our approach outperforms the state-of-the-art approaches. Our approach also shows good results with forensic sketches. The limitation with this dataset is its very small size. By increasing the training dataset, the accuracy of our approach increases, as it was demonstrated by our experiments.