Study On Liveness Detection Methods In Face Recognition

With the wide application of face recognition in life,the method of living detection in face recognition has attracted people's attention.The current living detection technology is mainly faced with attacks from fake pictures and fake video in practical applications.The safety and robustness of live detection techniques need to be improved.The efficiency of living detection needs to be greatly improved with the acceleration of face recognition detection.The main research is based on the posterior probability model of the blink detection to eliminate the photo attack and then we can eliminate the fake video attack by extracting SEMB-LBP features in face area and training and classification of sparse support vector machines.In this paper,the main work is illustrated as follows:1.The preprocessing of the acquired face images mainly includes histogram equalization,median filter denoising and illumination compensation.The pretreatment function is to make the results of the following detection of face area more accurate.Then the skin color model was established to detect the area of the face and locate it.Then use the posterior probability model to segment the face.Then,the facial features of the facial features of the facial features of three-court five-eye are used to locate facial features area and obtained the picture of human eyes and the five senses after the segmentation.2.Images attack is one of the live test of the main means of attack.In this paper,the morphological operation of the human eye area obtained by the posterior probability deformation model is presented,the total number of pixels in the iris of the human eye region is calculated.If it's a real living person or the human eye in a hypocritical video,the inner pixel of the iris will have periodic fluctuations and the number of blinks should be at least 1.But the blink of an eye occurs about 0 times in photos so the blink test can remove the photo attack.3.In this paper,extracting the 12 small sections of the human face area when facing video attack and the method of feature extraction is improved LBP feature extraction.The improved SEMB-LBP feature extraction method makes up the localization of the original LBP.So that the image texture information can be carefully described,and the overall information of the image is enhanced.Finally,the algorithm also decrease dimensions in order to improve the speed of feature extraction.Then,the semb-lbp features extracted from the image are used to train and classify the support vector machine.The advantage of the sparse support vector compared to the original support vector machine is that its solution has better sparsity and improves the operation rate and accuracy.