Research And Implementation Of Restoration And Recognition Of Locally Occluded Face Images

With the rapid development of artificial intelligence,facial recognition technology has been widely used in many fields such as security monitoring and facial payment.However,in practical scenarios,the accuracy of facial recognition technology faces many challenges,such as interference from factors such as lighting,occlusion,and angles,especially when facial occlusion is significant,the recognition effect is usually not satisfactory.To address these issues,this thesis proposes a research approach of repairing before recognition,which introduces a multi-stage facial restoration model based on local and global refinement and an improved facial recognition model based on the center loss function.The restoration model can effectively handle local occlusion,restore it to a high-quality unobstructed facial image,and input the repaired image into the improved recognition model to achieve good recognition results.The main research contents of this thesis are as follows:1.Aiming at the problem of occluded face restoration,this thesis proposes a multistage face restoration model,which divides the restoration work into three stages.First,a partial convolution mechanism is used to roughly restore face images with local occlusion,effectively reducing the interference introduced by occlusion areas.Then,a network with a smaller receptive field is used to repair local details,reducing the negative impact brought by distant unrelated areas.Finally,a network based on context attention mechanism is used for global refinement to efficiently establish the connection between occluded and non-occluded areas.The results of the experiment indicate that the suggested approach exhibits satisfactory performance.2.This thesis introduces a face recognition algorithm that utilizes the center loss function to address the challenges associated with face recognition..Utilizing deep neural networks for feature extraction and jointly constraining the center loss function and softmax loss function can enhance the intra-class cohesion and inter-class separation of facial features.Finally,experiments show that the proposed model performs well in recognition tasks.3.The development of a background system for restoring and recognizing occluded face images has been carried out.By demonstrating the functions of the system,this thesis shows that the system has good effects on restoring and recognizing occluded face images,and has practicality.