Facial Sketch-to-Photo Matching and Face Recognition using Local Invariant Features

In this paper, a proposed model is used to identify face sketch images based on local invariant features. Our system has three phases: (1) Feature extraction from the photos and sketches; (2) Reducing the dimension of feature vectors; (3) Matching the features of the unknown sketches with the features of the original photos. In this model, four local invariant feature extraction methods are used to extract local features from photos and sketches namely; Scale Invariant Feature Transform (SIFT), Speed Up Robust Features (SURF), Local Binary Patterns (LBP), and Weber Local Descriptor (WLD). Due to high dimensional features of all feature extraction methods, Direct Linear Discriminant Analysis (DirectLDA) is used. Moreover, in the classification phase, AdaBoost classifier is used. CHUK face sketch database images were used in our experiments. The experimental results proved that local invariant features achieved high accuracy and SIFT method achieved the best accuracy. Moreover, the effect of the different parameters were discussed and tuned to extract robust and discriminative features.