Study On Face Spoof Detection Based On Image Quality Analysis

To guarantee the security of face recognition system and prevent criminals from forging identity information to gain access to face recognition system,face spoof detection technology has attracted increasing attention.At present,most face spoof detection algorithms focus on analyzing the luminance information of face images and ignore the chrominance information of the image,so that the accuracy and stability of the algorithnm is disappointing in different application environments.There is a difference in color distribution between the real face image and the spoof image,so we can distinguish whether it is a real face by analyzing the chromaticity information of the image.In real-world applications,face spoof detection is performed in opening environments,thus unknown situations and unknown attack environments will be faced.To improve the generalization of the model,we conducted cross-dataset experiments.Due to differences in location and equipments,existing methods have a high error rate when conducting cross-database experiments.Research shows that illumination has the greatest impact on face spoof detection.Therefore,this thesis considers using illumination processing methods to remove impact of illumination factors in crossdatabase face spoof detection.In response to the above problems,this thesis gives the corresponding solutions and has achieved the following results:1.A face spoof detection method based on multi-feature fusion of colour space is discussed.Firstly,an RGB image is converted to HSV and YCb Cr colour space,respectively.Then uniformed rotation-invariant local binary pattern(LBP)features and colour moment features are extracted to describe the color texture and chromaticity changes on each colour channel of the image.At last,features are fed into an SVM classification and achieve results.This thesis designs multiple sets of controlled experiments to verify the rationality of the method,and obtain a high accuracy rate on the public data set.Compared with current traditional methods,our method demonstrates a low feature dimension and shows stable experimental performance.2.A face spoof detection method based on illumination analysis which can distinguish between real and fake faces in cross-database evaluation by removing illumination components of the image is discussed.Illumination pre-processing methods(e.g.logarithmic transformation,gamma correction and histogram equalizationand)and the method of extracting illumination invariants are used to eliminate the influence of illumination on face spoof detection.Experiments are conducted on two commonly used datasets.And the results are compared with other face spoof detection methods.The experimental results show that this method can achieve good results in cross-database experiments.