May 24th, 2002

Title : Elastica Inpainting,
Landmark based Inpainting from Multiple View
and Error Analysis

Sung Ha Kang
UCLA, Math dep

Inpainting problem arise when we have damaged old photo or art pictures which we need to revise. From the variational TV inpainting model suggested by Chan and Shen, we have improved the results by adding a Curvature term to the regularization functional. In addition, we consider more global inpainting problems in which the missing region are so large that most local inpainting methods fail. As an alternative to the local principle, we assume that there are other images with related global information to enable a reasonable inpainting. As one possible method, we propose using Landmark matching, interpolation, and inpainting. Finally, we study how the inpainting error depends on the shape and size of the inpainting domain. In particular, we first investigate two cases, first assuming image space is smooth, using Harmonic inpainting and secondly for binary image with TV inpainting.

This is collection of joint work with Tony. F Chan, Jianhong Shen, and Stefano Soatto.