Sparse Representation with Regularization Term for Face Recognition
In recent years there has been a growing interest in the study of sparse representation based classification (SRC) which has obtained great success in face recognition. However, SRC is overly dependent on the size of training samples while overlooking the
- PDF / 1,109,483 Bytes
- 11 Pages / 439.37 x 666.14 pts Page_size
- 38 Downloads / 259 Views
stract. In recent years there has been a growing interest in the study of sparse representation based classification (SRC) which has obtained great success in face recognition. However, SRC is overly dependent on the size of training samples while overlooking the correlation information that is critical to the real-world face recognition problems. Besides, some method considers the correlation information but overlooks the discriminating ability of sparsity. In this paper, we propose a new method called trace norm sparse representation based classification (TSRC) which introduces a regularization term in the SRC model and considers both sparsity and correlation. The TSRC method can benefits from both
1 -norm and 2 -norm, which is flexible and can obtain satisfacto-
ry results. Experimental results on 2 face databases clearly show that the proposed TSRC method outperforms many state-of-the-art face recognition methods. Keywords: Face recognition · Trace norm · Sparse representation based classification · Sparsity and correlation
1
Introduction
Face recognition, as one of the most successful applications of image analysis and understanding, has recently received significant attention and adequate development, especially during the past decades.Nevertheless, due to the different interference of different conditions cause corruption and errors of different degrees, for example, various facial expression, pose and illumination conditions, the face image processing effect is not so ideal. Furthermore, when the feature space is not sufficient sample database and high dimension, the existence of these problems will meet more challenges in face recognition. The conventional method of face recognition(sparse PCA [1], 2DPCA [2]) selects a limited subset or model from training samples, instead of the entire training set for image detection or signal classification and representation. So when the train sample space is small, the performances are not very good. These methods based on feature space, such as NN (Nearest Neighbor) and the support vector machine (SVM), when the image between different classes is very similar, will have a low recognition effect. © Springer-Verlag Berlin Heidelberg 2015 H. Zha et al. (Eds.): CCCV 2015, Part II, CCIS 547, pp. 10–20, 2015. DOI: 10.1007/978-3-662-48570-5_2
Sparse Representation with Regularization Term for Face Recognition
11
Therefore, face recognition methods based sparse representation classification emerge as the times requirement [3-5].Sparse representation method is based on the hypothesis that the testing images are approximation in a low dimension subspace which is obtained by the training samples, and then can be represented by a small number of training samples. Sparse representation based classification [3] (SRC) seek sparse representation of a query image in an over-complete dictionary, and then obtain recognition performance through comparing the minimal sparse error to identity the query image class. SRC can be seen as a generalization of NN and NFS, but it can ge
Data Loading...
