Research and Implementation of Facial Recognition Simulation Technology Based on Deep Learning
Low resolution facial images have poor feature saliency,resulting in low accuracy and weak stability in small target facial expression recognition.To solve the above problems,based on the MPII human detection dataset,a small target facial expression recognition algorithm,OPE-SVM algorithm,is constructed by integrating human keypoint detection and feature extraction through digital image optimization processing.The algorithm first uses the improved Open Pose deep learning algorithm to calculate the confidence map and affinity map of the key points of the human body,and uses the maximum suppression method to extract the single skeleton point map;Then by defining the base axis,the"nose"key points are used to correct the face skeleton map,and the face segmentation method is used to scale and segment the image to improve the accuracy of facial expression recognition;Then,the 3D LBP features and dynamic texture features are extracted and fused to improve the computational efficiency of the model.Finally,the SVM model for facial expression recognition is trained and optimized by using the ten-fold cross method.The results of ablation experiments show that different optimization structures in OPE-SVM algorithm have a positive impact on facial expression recognition results,and when the three groups of optimization structures are superimposed together,the model performance reaches its peak;The experimental results show that compared with other overlay models,the stabilityandaccuracy of OPE-SVM model are improved by 4.22％ and 4.14％,respectively,and the OPE-SVM model has better timeliness.To sum up,OPE-SVM algorithm can effectively improve the accuracy and stability of facial expression recognition of small targets,which has important simulation significance.
NETL
NSTL
万方数据
