Single-sample Face Recognition Based On Transfer Learning

Single-sample face recognition has a certain degree of difficulty and challenge. At present, the single-sample face recognition method has the following two problems: (a) The recognition rate is low. (b) the amount of calculation, data volume, the model is complex. So a single sample recognition performance can not be the best.To solve the above problems, this paper introduces the concept of transfer learning into single-sample face recognition based on the sparse representation classification framework,and makes a lot of experiments for other researchers to provide a reference for solving the problem of single sample using transfer learning and sparse representation classification framework. The main contents are as follows:(1) Based on the research background and significance of the research, this paper summarizes the history and present situation of the research at home and abroad, and summarizes the classic algorithms and points out the shortcomings of the existing algorithms.(2) The theory of transfer learning is introduced in detail, including the basic concepts,and the examples of transfer learning in life. The history of transfer learning is introduced,and the existing algorithms are briefly reviewed. The application of transfer learning and the future development direction are also discussed. This paper introduces the theory of transfer learning into single sample face recognition. The transfer learning algorithm which combines transfer learning with adaboost algorithm is studied, and the principle and steps of the algorithm are expatiated. It provides a guide for the researchers to understand the concept of transfer learning and tradaboost algorithm, which paves the way for the introduction of transfer learning.(3) The problem of single-face recognition is transformed into multi-sample face recognition based on the research of sample extension method. Combined with sparse representation classification recognition method, the recognition accuracy of single sample is improved. In the study of sample expansion methods, this paper expounds and contrasts several basic sample expansion methods, such as image transformation, sliding window and other expansion methods. The fifth chapter and the introduction of migration learning contrast experiment after paving the way.(4) The application of Sparse Representation Classification (SRC) algorithm and Extended Sparse Representation Classification (ESRC) algorithm in face recognition is studied. The classical sparse representation classification framework and the extended sparse representation classification algorithm are introduced. The accuracy of the classical SRC and ESRC algorithm for face recognition and the characteristics of the algorithm are compared and evaluated. Which is of guiding significance to the follow-up comparison experiment.(5) In this paper, we combine the transfer learning theory with the sparse representation classification method into a single sample face recognition problem to obtain a sparse migration dictionary which is helpful for classification from the auxiliary sample class,apply it to single sample face recognition problem,,The whole algorithm is based on the sparse representation of the classification framework, so the typical sparse decomposition algorithm is also analyzed. Through the experimental study on the two-dimensional image signal, compared and analyzed the Newton interior truncation method, homotopy algorithm,Performance of Long - day Algorithm. At last, the experiment is compared with the best homotopy algorithm. The feasibility of the migration learning algorithm is proved, and the shortcomings of the algorithm are put forward, and the improvement direction of the next step is expounded.