Partial Occlusion Face Recognition Based On CNN And Improved LBP

Face recognition in a controlled environment has achieved good results at this stage,reaching a recognition rate far beyond the human eye.However,facial recognition in uncontrolled and natural environments is susceptible to posture,illumination,occlusion,etc.,which reduces the performance of the algorithm,and the commercial application of some practical applications is not satisfactory for these interference problems.Among them,occlusion face recognition is one of the problems that must be solved at present.In this paper,the following two methods are used to study the face recognition under partial occlusion:(1)A feature of feature extraction based on gradient histogram of multi-scale landmarks is proposed,and the PCA whitening space is used for classification and recognition.First,we use the DLIB library detector to detect the face area in the image and perform face alignment based on the feature point position.After the facial image is obtained,Gaussian difference(DoG)is used as a band pass filter for image preprocessing.A multi-scale landmark gradient histogram(HOG)is used on the processed image to maintain the location and variation of the feature points to make the feature extraction more robust under face changes.The extracted facial features are classified in the PCA whitening space,and the similarity is calculated according to the Euclidean distance of the feature vectors in the space,thereby classifying the occluded faces.It can be seen from the experiment that the method has some improvement on the recognition of partial occlusion of the face,and is more robust to illumination and occlusion.(2)In face recognition,if the facial features are occluded,it becomes more difficult to identify the overall features of the faces in the image as the masked area increases,and the effective features that can be extracted are more and more To solve this problem,this paper uses the improved LBP algorithm for feature extraction based on face segmentation.This paper first uses the Faster-RCNN for face detection to get the face in the image,and then uses the TCDCN network to locate the left and right eyes,nose,left and right corners of the face obtained in the image.On this basis,the face is segmented.The improved LBP algorithm is used to extract the features of the face image of the block.After some feature extraction is completed,the occlusion is determined,and then the feature fusion is used to form the face for classification and recognition.It is verified that this method can greatly improve the performance of partial occlusion face recognition on the AR database.Figure[46]table[7]reference[50]...