Model-based enhancement of moving facial images

Ping, Xiaomeng (1998) Model-based enhancement of moving facial images. Masters thesis, Memorial University of Newfoundland.

[English] PDF - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. Download (5MB)

Abstract

This thesis investigates the application of 2-D object modeling to image enhancement and restoration. In particular, the topic of the thesis is focused on the recovery of accurate facial images from noisy and blurred videos. An immediate application of this is the improvement of security videos for identifying people, but there are other contexts, such as film restoration, where views typically fixate on faces, and there is therefore reason to give special importance to improving that part of the image. -- When several video frames of a static scene are available, noise reduction can be accomplished simply by averaging all the frames. In practice, movement of the objects (or the camera) often prevents this. For example, a security video will probably provide a frame sequence of an individual in motion. Our strategy for enhancing facial images is to employ warping to eliminate these variations and to produce a uniform image sequence which will be involved in frame averaging. -- A 2-D face model which consists of a set of landmarks that describe the main facial feature points is used in the thesis. The face model is matched frame-by-frame to the sequence by a human operator. Given the landmarks, warping is performed to transform all the noisy and blurred frames to the same target frame. Then temporal averaging among the transformed resulting frames is applied to produce an enhanced result. Finally, spatial filtering for blur removal is optionally applied. -- To test the effectiveness of this strategy, some experiments are designed and performed. The results demonstrate that the supervised object-based temporal filtering strategy is simple and effective for enhancement of faces in videos.

Actions (login required)

View Item