Disguised Face Recognition by Using Local Phase Quantization and Singular Value Decomposition
الموضوعات : Journal of Computer & Robotics
Fatemeh Jafari
1
(Faculty of Computer and Information Technology Engineering, Qazvin Branch, Islamic Azad University, Qazvin, Iran)
Hamidreza Rashidy Kanan
2
(Department of Electrical, Biomedical and Mechatronic Engineering, Qazvin Branch, Islamic Azad University, Qazvin, Iran)
الکلمات المفتاحية: Disguised Face Recognition, Local Phase Quantization, singular value decomposition, Fourier transform, Manhattan and Euclidean Distance,
ملخص المقالة :
Disguised face recognition is a major challenge in the field of face recognition which has been taken less attention. Therefore, in this paper a disguised face recognition algorithm based on Local Phase Quantization (LPQ) method and Singular Value Decomposition (SVD) is presented which deals with two main challenges. The first challenge is when an individual intentionally alters the appearance by using disguise accessories, and the second one is when gallery images are limited for recognition. LPQ has been used for extraction of the statistical feature of the phase in windows with different sizes for each pixel of the image. SVD is used to cope with the challenge of the gallery images limitation and also with the help of original images extracted from that, every single image turns to three renovated images. In this study, disguise is intended as a blur in the image and Local phase quantization method is robust against the disguised mode, due to the use of the statistical feature of the Fourier transform phase. Also the use of different-sized window instead of fixed window in feature extraction stage, the performance of the proposed method has increased. The distance of images from each other is computed by using Manhattan and Euclidean distance for recognition in the proposed method. The Performance of the proposed algorithm has been evaluated by using three series of experiments on two real and synthesized databases. The first test has been performed by evaluating all the possible combinations of the different-sized windows created in the feature extraction stage, and the second experiment has been done by reducing the number of gallery images and then the third one has been carried out in different disguise. In all cases, the proposed method is competitive with to several existing well-known algorithms and when there is only an image of the person it even outperforms them.
