Research And System Implementation Of Face Recognition For Local Occlusion

With the development of science and technology,the application of face recognition in life is becoming more and more extensive.Compared with other recognition technologies,face recognition has the characteristics of non-contact and strong ability to track afterwards.However,when the face is occluded,face recognition will face severe challenges,and the accuracy of face recognition will be greatly affected.Therefore,it is of great practical significance and research value to study the face with occlusion in face recognition.According to the different types of face occlusion,this paper conducts an in-depth study on the face recognition method of eyeglasses occlusion and the partial occlusion face recognition method where the occlusion position,area,and shape are unknown.The specific research content are as follows:(1)Aiming at the problem of facial glasses occlusion,this paper proposes a two-dimensional face recognition method based on improved total variation model to remove eyeglasses occlusion.First,detect the attributes of eyeglasses in the face image.The eyeglasses attribute detection includes four types:the position of the frame,the type of the frame,the color of the lens,and the reflection area.Then,according to the detection results of the eyeglasses attributes,the method of improving the total variation model is used to remove the occlusion of the face eyeglasses.Finally,for the face after removing the eyeglasses,the LBP algorithm is used to extract facial features and perform face recognition.Experiments prove that the eyeglasses removal effect based on the improved total variation model proposed in this paper is better than other algorithms,and the face image after eyeglasses removal by this algorithm has a higher recognition rate compared with other algorithms.(2)Aiming at the problem of face occlusion with uncertain occlusion position,area and shape,this paper proposes a three-dimensional partial occlusion face recognition method based on multi-dictionary weighted cooperation.First,take corresponding preprocessing operations on the face and partition the face.Second,use the curvature features of the three-dimensional image to detect multiple key points in each partition.Then,based on key points,a variety of local features are extracted to form feature descriptors.Finally,a feature dictionary is constructed for the feature descriptors of each partition,and weighting items are introduced to weaken the weight of points with larger key points error during recognition.The weighted cooperation of multiple feature dictionaries is used to complete face recognition.Experimental results show that compared with other methods,the proposed 3D local occlusion face recognition method based on multi-dictionary weighted cooperation has better stability and recognition rate.(3)Designed and implemented a face recognition system for partial occlusion.The system implements a two-dimensional face recognition method based on the improved total variation model to remove the occlusion of eyeglasses and a three-dimensional partial occlusion face recognition method based on multi-dictionary weighted cooperation.Whether it is a two-dimensional face occluded by glasses or a three-dimensional partial occlusion of a human face where the location,area,and shape of the occlusion are uncertain,the system can maintain a good recognition effect.