Face hallucination method via eigenfaces estimation and Markov high-frequency compensation

This paper presents a method of turning a low-resolution face image to its corresponding high-resolution one, namely face hallucination. Our method can be roughly separated into two steps: Firstly, we employ a greedy algorithm Matching Pursuit (MP) that is widely used in sparse signal processing to generate an intermediate face image which captures holistic information. This step is with a sample image set containing a lot of high-resolution face images and their corresponding low-resolution ones. Secondly, we take advantage of Markov model to compensate high-frequency information which is local information but badly represented in the previous step. We present a simplified but efficient compatibility function form with fewer computation loads. A series of experiments from both subjective and objective aspects demonstrates that our method is able to achieve satisfactory results.