Linear discriminant analysis representation and CRC representation for image classification
Chunwei Tian, Qi Zhang, Guanglu Sun, Yuan Liang Sun · 2016
Linear discriminant analysis (LDA) has good stability and applicability for real-world tasks, because it can use orthogonal vectors to extract suitable features, which not only suppresses the difference between different classes and unrelated identification information, but also is not sensitive to illuminations and varying facial expressions. LDA can also decrease the dimension of original images, which improves the efficiency of recognition images. This paper proposes the simultaneous use of LDA and collaborative representation classification (CRC) for classification of images and obtains excellent performance. LDA is used to extract features and construct virtual images. This proposed method can automatically and quickly extract suitable features without any manual setting. The obtained virtual images are also different and complementary with original images, which effectively improve the performance of image classification. This novel method is not only simple and easy to implement, but also doesn't have any parameter. As a consequence, it has good prospect of practical applications. To fully verify the performance, we design comparative experiments on face datasets.