Research On Face Spoofing Detection Based On Improved LBP

Along with the rapid development of internet technology,the application of face recognition has been widely used in daily life,such as the authentication functions on mobile devices and the security systems in airport or station.Face recognition technology not only improves the efficiency of identity verification greatly,but also brings a quick and convenient life experience for users.Nevertheless,spoofing attack will occur when the deceiver fakes user's face and through the facial recognition system,it is a serious threat to the security of user's property and information.The spoofing detection technology in existing applications needs the assistance of human-computer interaction,which will bring a worse experience for customers.Therefore,it is more than meaningful to solve the risk of system before face recognition without human-computer interaction.Due to LBP is simple,efficient,and has better generalization ability in the database of complex samples,it has become the mainstream method of feature extraction in face spoofing detection field.But,the LBP feature is relatively simple and cannot describe the texture details and spatial structure of the image.Given this,we will explore face fraud detection technology from refining LBP features,extracting texture spatial structure information,and transforming image representation.Particularly,the main contributions of the dissertation and our innovations are as follows:(1)Face anti-spoofing algorithm based on block color MB?LBP texture.The noise signal,video artifact,print defects and other fraud clues will be produced during the fraud image manufacture,but the fraud clues are not conspicuous in high resolution image.Actually,the live image is different from its reconstruction,since the gamut of images from different devices is inconsistent.Consequently,we select a specific color space to distinguish the spoofing face from real and extract large-scale local texture features by using MB?LBP,so as to capture sample characteristics in complex datasets.Additionally,the feature image is segmented to extract its spatial structure information.(2)Research on face anti-spoofing algorithm based on DQ?LBP.LBP is widely used in face fraud detection currently because of its simplicity and efficiency.However,LBP only considers the signal of the center point and its neighbors.For print attack and video attack,we propose a difference quantization local binary pattern(DQ?LBP)approach for refining the feature of traditional local binary pattern(LBP)by quantifying the difference between the value of central pixel and its neighborhood pixels.DQ?LBP can extract the difference information between the local pixels without increasing the original dimension of LBP,and thus be able to describe the local texture features of images more accurately.In addition,we use the spatial pyramid(SP)algorithm to calculate the histogram of DQ?LBP features in different color spaces and cascade them into a unified feature vector,so as to obtain more elaborate local color texture information and spatial structure information from the face sample.Consequently,the performance of fraud face detection algorithm in this paper has been further improved.(3)Face spoofing detection based on chromatic ED-LBP texture feature.As we all know,noise signal is an important clue for face anti-spoofing detection,but it will be generated inevitably during the acquisition of real face.Although noise signals of high-quality live samples are difficult to capture,real samples are easily misclassified in data sets which composed of complex samples.So,we equalize the local pixel to reduce noise signal,and combine it with DQ?LBP to form the equilibrium difference localbinary pattern(ED-LBP).In addition,we propose two kinds of ED-LBP which named ED-LBP('(10)')and ED-LBP('x ').The proposed models not only have high detection performance but also reduce the characteristic dimension to 1/16 of the ED-LBP.