This algorithm applies the concept of second-order tensor to the original canonical correlation analysis method, which effectively avoids the singularity of covariance matrix and greatly reduces the computational complexity.

Related papers:

Research on Linear Subspace Face Recognition Algorithm and Attitude Problem

Huang Likun

In the past thirty years, automatic face recognition technology has attracted great interest and extensive attention from researchers in various related fields, spanning many research fields such as image processing, computer vision, neuroscience, statistics, pattern recognition and so on. With the development of science and technology, automatic face recognition technology is gradually applied in the fields of commerce and public safety. From the static matching of face photos in controllable formats such as passport, ID card and driver's license to face image recognition in real-time monitoring, automatic face recognition technology has played a huge application value. At present, when the face images used for recognition are obtained under controllable conditions, such as clear front and suitable lighting conditions, the recognition rate has reached an acceptable level. However, when the user does not cooperate and the conditions are not suitable, such as the non-frontal, low pixel and poor lighting conditions intercepted in the surveillance video, the recognition rate will be greatly reduced, and in some cases the recognition rate is even less than 30%. Therefore, we can see that there are many problems in face recognition technology that have not been solved until now, and these problems are the real problems that face recognition technology urgently needs to solve in daily life. This paper mainly studies two problems in face recognition technology: face recognition algorithm based on linear subspace learning and face recognition algorithm under posture change. Firstly, this paper expounds the face recognition algorithm based on linear subspace learning, and introduces principal component analysis, linear discriminant analysis and canonical correlation analysis in detail. Based on the analysis of linear subspace face recognition algorithm, a new face recognition algorithm based on linear subspace learning (two-dimensional discriminant canonical correlation analysis) is proposed. This algorithm applies the concept of second-order tensor to the original canonical correlation analysis method, which effectively avoids the singularity of covariance matrix and greatly reduces the computational complexity. Face image recognition under posture change is one of the main problems in automatic face recognition research. This paper summarizes the face recognition algorithms under different posture changes. Because the transformation of face image under posture change is nonlinear, the face recognition algorithm based on sub-region can solve the face recognition problem under posture change well. Based on local linear regression method and Gaussian probability model, a new face recognition algorithm based on sub-region attitude change (weighted sub-region similarity face recognition algorithm) is proposed. The algorithm uses local linear regression method to generate virtual face frontal image.