Research On The Quality Assessment Of Bayonet Face Image Based On Multi-feature Fusion

Face quality assessment is a method for obtaining facial image quality scores by performing feature analysis and research on face images.In the face recognition system of bayonet,a large number of low-quality face images are received which provided a problem that the recognition accuracy is reduced and the calculation load is high.However,Face quality assessment is a key step in face recognition,Face quality assessment is a key step in face recognition.It can provide high-quality facial images for recognition technology to improve the accuracy and efficiency of the recognition system.To the end,this paper focuses on the quality assessment of bayonet face image based on multi-feature fusion.The main research is as follows:1.Focusing on the characteristics of bayonet face affected by light and the change of viewing angle,a face quality assessment method based on multi-feature fusion was proposed.Firstly,a double-layer evaluation framework is applied to the texture features of multiple faces,and the weights of feature fusion are trained by the Newton optimization method to obtain the texture quality feature fusion face quality assessment algorithm.Then select a variety of apparent correlation features such as HOG(Histogram of Gradient),GIST(Generalized Search Trees),and GABOR to perform combined tests to determine the optimal face quality assessment algorithm.Finally,the evaluation algorithm is combined with the face recognition module of SeetaFace to constitute a face recognition method,and different scoring thresholds are set to test the recognition performance.Experiments show that the face image selected by the face quality assessment algorithm based on HOG-GIST feature fusion can effectively improve the accuracy of face recognition.2.In view of the strong expression ability of face key points and external contours,this paper proposes a feature fusion method based on key points of human faces,which is used to evaluate the quality of face images.The method includes firstly extracting MB-LBP(Multiscale Block Local Binary Pattern)features according to the coordinates of image sparse feature points,and then cutting out the human face contours based on sparse feature points to extract Hog and Gist features.Finally,a double-layer evaluation framework is adopted for the three features,and the feature fusion weights are trained by the conjugate gradient method to obtain the texture feature fusion face quality assessment algorithm.Secondly,face key image is calibrated with dense key points.MB-LBP features are extracted from key points using a single-layer evaluation framework.The feature weights are trained by conjugate gradient method to obtain face quality assessment algorithm.Two kinds of evaluation algorithms are combined with SeetaFace's face recognition module to form a face recognition method,and different scoring thresholds are set to test the recognition performance.The comparison experiment shows that the face quality assessment algorithm based on the dense key points makes an objective score on the face image,and the selected face image can effectively enhance the accuracy of face recognition.In summary,in order to improve the accuracy of the bayonet face recognition system and reduce the computational load of the system.This paper proposes a face quality assessment method based on multi-feature fusion.By analyzing the effect of global texture and face feature points on image quality assessment,a reliable assessment scheme was determined and the reliability of the image quality assessment method was verified in a face recognition system.