Face Recognition Based On Fractional Fourier Transform

Face recognition from still images and video sequence has been an active research area due to both its scientific challenges and wide range of potential applications such as biometric identity authentication, human-computer interaction, and video surveillance. Within the past two decades, numerous face recognition algorithms have been proposed as reviewed in the literature survey. Even though we human beings can detect and identify faces in a cluttered scene with little effort, building an automated system that accomplishes such objective is very challenging. The challenges mainly come from the large variations in the visual stimulus due to illumination conditions, viewing directions, facial expressions, aging, and disguises such as facial hair, glasses, or cosmetics. As the generalized form of the Fourier transform, the Fractional Fourier transform (FrFT) can be interpreted as a rotation of signals in the time-frequency plane. The analyzed signals can be mapped to any domain between time domain and frequency domain with the method of FrFT to extract images features more effectively. Fractional Fourier Transforms of different orders of the face images, corresponding to different fractional domain space, can be more effective to extract the features to characterize the face images. Therefore, the FrFT not only can achieve the functions as the traditional theories, but also be more universal and flexible. As a result, the FrFT is hoped to obtain better results in the field of facial face recognition than the traditional theories. This paper studied the properties of the fractional domain of the face images, and proposed the feature extraction methods, the main work and contributions of the paper are as follows:1. In face recognition, there are great challenges with variations arising from illumination, expression and other factors. Since the fractional Fourier transform feature is robust to illumination and expression variations and has been used in face recognition area, we propose a novel algorithm to face recognition with the local region histogram of the two dimensional fractional Fourier transfonn (2D-FrFT) magnitude and phase (LFMP). Specially, a face image is modeled as a "histogram sequence" by concatenating the histogram of all local regions of 2D-FrFT magnitude and phase binary pattern maps. The histogram intersection is used to measure the similarity of different LFMP binary pattern maps and the nearest neighborhood is exploited for final classification. Finally, the genetic algorithm is introduced to select appropriate value of transform order for discrete fractional Fourier transform. We evaluate our approach on ORL and FERET face databases. Extensive experimental results verify the effectiveness of our LFMP descriptor.2. This paper presents a novel Multiple Order discrete fractional Fourier Features (MOFF) method based on Sparse Principal Component Analysis (SPCA) for face recognition. The fractional Fourier transform (FrFT) has been used to image processing with its robust to illumination and expression. Specially, the magnitude of FrFT, whose energy displays constringent characteristic, is handled by SPCA to further divide into the main energy of magnitude part (MMP) and the remaining energy of magnitude part (RMP), which are combined into the hybrid magnitude part (HMP) to fuse complementary features. Then for fractional Fourier Features with individual transform order, the hybrid fractional Fourier features (HFFF) is formed and consists of three fractional Fourier features:HMP, real part (RP) and imaginary part (IP). Finally, the HFFF generated using three fractional Fourier features with different transform orders is fused by means of the weighted summation rule-the decision level fusion-to derive the MOFF for face recognition.In addition, the Greedy search is introduced to select the transform order of the FrFT. Experimental results of MOFF on the ORL and AR databases verify the effectiveness of the results by using these new modifications3. Sparse multimodal biometrics recognition (SMBR) has achieved robust face recognition results against occlusion, by introducing an identity occlusion dictionary to sparsely code the occluded portions in face images. However, occlusion dictionary must be orthometric and setting identity occlusion dictionary is a strict condition. As the robustness of 2D-FrFT to noise of face images, a joint local sparse representation combining with robust principal component analysis (JLSRPCA) based on 2D-FRFT is presented for face recognition on occluded images. Specially, candidate regions randomly chosen from integrated images are used to build sparse representation, combining with robust PCA to search the best occlusion matrix to weaken influence of bad elements (occlusion, illumination, expression etc.). Besides, a novel maximal similarity based patches choosing mechanism is also described to select effective patches. Moreover, each of patches is weighed by a proposed patch quality measure. Extensive experiments on both Yale B and AR datasets have shown the effectiveness of the proposed method.