Abstract
Face hallucination is of great importance in many applications. In this paper, a novel two-stage framework is proposed for hallucinating high-resolution (HR) face image from the given low-resolution (LR) one. In contrast to the existing methods, where the finer details are ignored, our framework pays more attention to the further local details enhancement. In the first stage, the local position-patch-based method with locality constraint is introduced to obtain the initial estimate image. In order to generate more reasonable face image and reduce noise, our method only represents the input LR patches over the similar training patches in the same position. In the second stage, the initial estimate image rather than residual image is directly used as the input to obtain the final HR image via local position-patch-based method. Besides, contextual information of position-patch is taken into consideration to generate more precise details in the second stage. Extensive experiments on the open face database illustrate that the proposed method achieves superior performance in comparison with state-of-the-art methods.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call