Multi-angle Trace Sampling Feature Extraction Algorithm And Its Application In Face Recognition

Face recognition technology is one of a popular research fields in computer vision.The main basis of its classification and recognition is to extract the features that accurately reflect the face information.Currently,face recognition technology has been very mature under ideal conditions,while in the real cases,facial recognition effect is not very good,the main influence factors is illumination,expression,shelter and multi-scale problems,Especially for the blurred face recognition caused by objective factors such as distance and movement in practical application,the effect still has a large space of improvement.Feature extraction is the most critical step in the process of face recognition.A good feature extraction method largely determines the accuracy of face recognition.Aiming at the problems in non-ideal environment and the low rate of blurred face recognition,this paper deeply analyzes the existing face feature extraction methods,and proposes the fusion feature extraction method of multi-angle trace on the basis of Trace transformation.The main contributions of this paper are as follows:1.This paper proposes a feature extraction method of local gradient pattern of multiangle trace.This method uses gradient instead of pixels to extract features from the image,which changes the feature extraction method of local gradient pattern,and sampling points are selected from the beginning to the end of each tracing line of the image in an "orderly" manner and coded.Meanwhile,the idea of image rotation is also introduced.A set of feature information is extracted from every rotation angle of the image,and finally features with "ordered" global structure are extracted,which enhances the expression ability of image feature space structure and rotation invariance.2.Aiming at the situation that gradient local binary pattern is limited to local texture feature extraction of image,a feature extraction method for gradient local binary pattern of multi-angle trace is proposed,which enables the face image features extracted with global structure.This method not only improves the extraction method of features of local binarypattern gradient,but also introduces the idea of image rotation in Trace transformation.Each rotation Angle of the image can be used to extract a group of feature information.3.This paper proposes a HOG feature extraction method of multi-angle trace,which changes the original HOG feature extraction method,which is conducted on each trace line to extract the texture information of each trace line,and also combined with the idea of image rotation in Trace transformation to extract the face feature information with global structure.To sum up,the fusion feature extraction method of multi-angle trace proposed in this paper has a strong generalization ability and can be applied to a variety of features.Meanwhile,experiments in standard face database show that this method achieves higher recognition rate than Trace transform method,LBP method and HOG method,and it is better than the traditional convolutional neural network method in blurred face database.It can be seen from the experimental results that the rotation matrix feature extraction method proposed in this paper has a strong application prospect in the field of face recognition.