Appearance-based face recognition using aggregated 2d gabor features

Face recognition performed in a controlled environment can be transformed to classical image retrieval/pattern recognition of frontal 2D images of a person, i.e. mug shot. Current holistic appearance based face recognition methods require a high dimensional feature space to attain fruitful performance. We, therefore, propose a relatively low feature dimension scheme to cope with the transformed face recognition problem. Aggregated Gabor filter responses is employed to represent face images. We have conducted experiments on two testing sets of a face image database. Each set contains over 3,000 images of the same 120 subjects and they differ from each other in the preprocessing of image size. We have compared the performance of using our method and PCA and the performance of using L1 and L2-norm as (dis)similarity measures to guide the retrieval process.