Face hallucination produces high-resolution facial images from low-resolution inputs. In this paper, we propose a contour-based face hallucination method. Since our goal is face recognition rather than visual effects, the contour information of facial-parts (such as eyes) is important. We focus on the eye parts and reconstruct eye contours instead of dividing them into small blocks. We obtain the eye contours by using the Active Appearance Model (AAM), and transform training images based on contours. We confirm that the proposed method significantly enhances face recognition performance.
抄録全体を表示