Research On Incremental Sparse Tensor Regression Algorithm For Face Recognition

The wide application of information makes identification have practical significance. Face recognition is one of the main technologies of identification. Because of its non-mandatory, non-contact and the characteristics of concurrency it has been a hot research of machine learning and pattern recognition.In this paper, we present an incremental sparse linear regression algorithm for face recognition based on linear regression classification algorithm in order to solve the problem of solving linear regression classification algorithm in large scale data. The main work is as follows:(1) An incremental linear regression classification algorithm is proposed. By using the matrix theory, the algorithm can be improved, and the learning speed of the model can be accelerated without affecting the accuracy of the model.(2) An incremental sparse linear regression classification algorithm for face recognition is proposed. On the basis of incremental linear regression classification algorithm, Lasso regression is used to sparse representation coefficients so that the model has higher recognition accuracy when the data contains a large amount of redundant information.(3) An incremental sparse tensor regression classification algorithm for face recognition is put forward. The incremental sparse linear regression for face recognition has been generalized to tensor data by using the knowledge of tensor so that the algorithm has better recognition accuracy and faster learning time for large-scale and high-dimension tensor data.(4) Experiments on the face database show that compared with linear regression classification algorithm, incremental sparse linear regression algorithm has faster training speed and higher accuracy of recognition. The greater size of the data has, the advantage of the algorithm is more obvious. Compared with vector model, tensor model has faster training speed and higher recognition accuracy on large-scale, high-dimension data. The larger size and the higher dimension of the data has, the more obvious the superiority of the tensor model is.