Inverse transformation based weighted fusion for face recognition

Abstract

Score fusion is a promising approach in face recognition. This paper proposes an algorithm that can integrate original images and virtual images. We obtain face-like virtual images that provide reverse representations of original face images by utilizing a new non-linear transformation. Pixels with high intensities in original images correspond to pixels with low intensities in virtual images and vice versa. The correlation coefficient between the two kinds of data sources is relatively low, which indicates that the reverse face images are complementary to original face images. In this study, a linear model is used for calculating the distances between test samples and training samples, and all the distances are sorted in ascending order; we exploit the difference between the best score and the second-best score to calculate the weight. According to the experimental results, the proposed method outperforms other state-of-the-art methods in recognition accuracy and has a high computational efficiency.