Weighted nearest neighbor algorithm and collaborative representation for image classification

The nearest neighbor classification method (NNCM) has good flexibility and applicability, because it can directly and quickly use the nearest distance between all the training samples and a test sample with unknown class label as similarity to classify the unknown test sample. So, NNCM has been widely applied in pattern recognition, image processing, data compression and network analysis, e. However, when samples are not evenly distributed, NNCM might have a high recognition error rate. In this paper, we propose to use the combination of weighted nearest neighbor algorithm (WNNA) and collaborative representation classification (CRC) to classify images, which has more satisfactory performance than general sparse classification algorithms. Our proposed novel method has the following main three phases. The first phase uses linear combination of all the training samples to obtain virtual images and renew to represent original image, which is more robust than the original image. The second phase utilizes CRC to respectively obtain scores of the original image and virtual image. The final phase uses different weights to integrate obtained scores and classifies images. The major contribution of the proposed method is that the obtained WNNA features are complementary with original face images. Our experiments show that the simultaneous use of WNNA and CRC obtains high accuracy for image recognition.