The Gordian Technique Research On Face Liveness Detection

Face recognition system is suffering photo spoofing, because we can login in a face recognitionsystem by using a face image of the valid user. Face image contains rich information, which is a complexpattern of machine visions. Digital face image contains a series of information, such as shape, texture andcontour which can be used to characterize the age, skin, identity and etc. For example, a color image, withwidth of240pix and height of320pix, has230×400dimensions. So it is abundant to reflect the faceinformation. With the continuous progress of science and technology, the rapid development of computerscience, pattern recognition has broke new ground for image forensics. Therefore, it is great significance toauthenticate whether the scene of the digital image is authentic.In this paper, according to the difference between real images and false photos, texture and statisticalfeature are extracted in different ways, several algorithms are proposed to identify whether the face in frontof the camera is alive or not. The algorithms the paper proposed only utilize a single image to identity. Themethod proposed in this paper has been further improved compared to the former method, and showed agood performance.First, discrete cosine transformation can be used to map the import face information to the lowfrequency of DCT domain. However, there are varieties of noise in the high frequency part of the image.Considering the noise factor can be eliminated by removing the high frequency part, we propose to use thelow-frequency coefficients to serve as feature, and the face samples will be used to classify the SVMclassifier.Second, texture is one of the most import features for image, which represents the gray leveldistribution of the pixel in the neighborhood. Texture analysis plays an important role in the image research.Texture is an effective way to characterize the difference between the live face and the impostor. So the twoclass photos can be taken apart by choosing the suitable texture descriptor..Compared with the otherdescriptors, The tamura texture and LAP texture we proposed achieved very good results.Third, Histogram of digital image plays an important role in the analysis and observation of the image.Histogram of an image describes the different color gradations of images, which reflects the statistical lawof the color, and it is also very useful in pattern recognition and image segmentation. After heavyresearches on the real face photos and fake reproductions of photos, we found the two kind of photos showdifferent distribution in histogram, so the algorithm based on HSV color histogram is put forward.Fourth, because the pixel gray values of the image can be used to form a matrix, the singular values ofmatrix can be used to describe the inherent attribute of the image. A number of experiments show thatdistinguishing ability are tremendous differences when the singular feature are extracted in different ways.This paper presents a reasonable method of feature extraction, which is used for in face liveness detection.