Face hallucination and recognition in social network services

Abstract

Due to the rapid growth of social network services such as Facebook and Twitter, incorporation of face recognition in these large-scale web services is attracting much attention in both academia and industry. The major problem in such applications is that visual data provided are in general noisy, low resolution, prone to degradation due to lighting and other adverse effects. This paper proposes a novel face hallucination method with an aim of improving face recognition performance with the photography in social media data, which is modeled as a progressive process which explores the specific face characteristics and priors based on frequency bands analysis. In the first stage of our algorithm, initially estimated middle-resolution images are generated based on a patch-based learning method in discrete cosine transformation (DCT) domain as the first-scale restoration image. According to the relationship between the high-resolution face images and their lower resolution ones, the DC coefficients and AC coefficients are estimated separately. In the second stage, aiming at generating more refined high-resolution face images, a DCT up-sample algorithm emphasizing on low-frequency bands preserving is applied. Meanwhile, an interpolation-based method is presented in spatial domain to obtain high-frequency bands. Extensive experiments show that the proposed algorithm effectively helps improving the recognition performance of low-resolution faces.