Multi-angle Face Recognition Research Based On GMRF

With the rapid development of artificial intelligence in the 21st century,intelligent prevention and control is a branch of many fields of artificial intelligence,and multiangle face recognition technology will make a qualitative leap in intelligent prevention and control.At present,most face recognition technologies are positive face recognition,but the faces captured in reality are face images with angles,therefore,multi-angle face recognition research has important significance for face recognition in the real scene.This paper focuses on how to extract face image features and improve the accuracy of face recognition under the condition of multiple angles.Specific research work is as follows:1.A Gaussian random field model(GMRF)based on image spatial domain is implemented to extract features from multi-angle face recognition.The calculation of Gausmarkov random field model is small.For feature extraction of multi-angle face image,there are some problems,such as imperfect information extraction,incomplete information extraction and low recognition rate.According to the practical characteristics of multi-angle face recognition technology,this paper extracts the GMRF features of multi-angle face image space domain for classification and recognition.At the same time,this paper also carries on the gradient extraction of the image,and uses GMRF to achieve the feature extraction of the gradient domain.GMRF can extract the transformation trend of image pixels,so that GMRF can scan the spatial image and the gradient domain image,so as to describe the multi-angle face information.The experimental results show that the nonlinear Gaussian Markov random field model is used to extract image texture features in the image domain and gradient domain,which can effectively improve the multi-angle face recognition rate.2.A GMRF feature extraction algorithm based on image frequency domain is proposed.In view of the deficiency of information expression and feature extraction in traditional image spatial domain features,the algorithm of GMRF feature extraction based on image frequency domain is proposed in this paper.Firstly,the spatial pixel transform of two-dimensional multi-angle images is converted to frequency domain by Fourier transform,and the distribution of image pixel values is converted to frequency domain distribution and processed.Secondly,GMRF features such as real part and imaginary part in frequency domain of multi-angle face image are extracted respectively,and the GMRF features of multi-angle face image in frequency domain are composed by different feature combinations.Experimental results Compared with the airspace,the multi-angle face recognition rate increased by 4.21%.3.A multi-angle facial feature combining GMRF Features and Speeded Up Robust Features is proposed.SURF is mainly used to represent local features of images.It inherits the advantages of scale invariance of SIFT and can well extract local detailed features of multi-angle face images.Its repeatability and specificity complement the details of multi-angle face features and fully meet the requirements of the subject.The experimental results show that the combination of features can effectively supplement the deficiency of GMRF features in face detail expression,and can effectively improve the multi-angle face recognition rate.In this paper,15 kinds of identification experiments with translation and tilt angles ranging from-90 degrees to+90 degrees were carried out on the Head Pose Image Database(HPID)dataset,as well as comparison experiments with other methods.The experimental results show that it has good recognition effect under 15 different angles.When nonlinear GMRF features in the frequency domain were combined with SURF features,the recognition rate reached the highest,reaching 99.57%.Finally,this paper also experiments on the method in the self-made multi-angle face database,and the recognition rate reaches 98.83%,which effectively proves the effectiveness of the proposed method.