Visible facial blemishes can be annoying and embarrassing in our daily lives. This is particularly true when blemishes are permanently captured on video. For film and broadcast production, blemishes on actors/actresses may not be acceptable and must often be removed by post-production processing. Such processing can be incredibly slow and tedious, requiring a digital artist to "retouch" the video footage frame-by-frame. For amateur video editing, this manual processing is impractical with existing tools....[ Read more ]

Visible facial blemishes can be annoying and embarrassing in our daily lives. This is particularly true when blemishes are permanently captured on video. For film and broadcast production, blemishes on actors/actresses may not be acceptable and must often be removed by post-production processing. Such processing can be incredibly slow and tedious, requiring a digital artist to "retouch" the video footage frame-by-frame. For amateur video editing, this manual processing is impractical with existing tools.

This thesis presents a unique framework, referred to as Video Cosmetics (VC), designed to remove unwanted blemishes from video footage with minimal user assistance. Using the VC framework a user needs only to specify the blemish in an initial frame; the system then automatically tracks and removes the blemishes from subsequent frames. This task is accomplished by combining feature tracking, pixel classification, and temporally coherent image inpainting into a single processing framework. The result is an efficient post-production tool that substantially reduces the amount of time and effort needed to remove blemishes from video footage.