Research On Face Recognition Algorithm With Occlusion Based On KPCANet

Face recognition technology is still facing some great challenges today.Faces acquired in natural environment are likely to be occluded and vulnerable to changes in lighting posture,This greatly reduces the recognition accuracy.Most existing depth learning models do not have the ability to locate occlusion and quickly identify it,and The partial depth learning model has poor nonlinear fitting ability.A new algorithm based on KPCANet for face recognition with occlusion is proposed.To solve the problem of light attitude change,and applying kernel function mapping to principal component analysis network model After extracting the shallow features of the image,the kernel transformation is carried out on PCANet to improve the non-linear fitting ability,Make the model more robust.Feature point detection is carried out on face samples,ensure face samples each key point detection and extraction,A classifier is designed to judge whether the face key points are occluded or not,the classifier can effectively distinguish between input facial image block belongs to a specific face point or to the background picture or key block with a shade.Then PCANet feature extraction,Input the extracted feature matrix into the classifier with lib library to train SVM model,and the different key points of training model combines characteristics of the SVM model group,combined with block discriminant classifier and different shade type using different sample prediction model group to complete the task of face recognition with occlusion,the experiment proves that The algorithm has achieved good results on common face occlusion types,Large area occlusion also has a certain recognition rate,It also has a relatively high recognition rate for occlusion caused by illumination and expression changes.