Perspective and Parallax Preserved Inpainting for Stereoscopic Architectural Scene Images

• Main Author: 陸品樺
• Other Authors: 林奕成
• Format: Others
• Language: zh-TW
• Published: 2011
• Online Access: http://ndltd.ncl.edu.tw/handle/19399998979522812353

碩士 === 國立交通大學 === 多媒體工程研究所 === 100 === Image completion algorithms are powerful for removing undesirable obstacle objects. Nevertheless, when the undesirable obstacle objects cover a large amount of the background and the background is artificial constructed, structure of the background is difficult to maintain. In this thesis, we present an automatic image completion method for stereoscopic architectural scene images. We take advantage of two-view information to reduce obstacle pixels and preserve parallax consistency. A vanishing point and vanishing line prediction is used to perspective correct the input images to eliminate perspective artifacts. A structure-enhanced patch searching algorithm is also proposed to preserve the architectural structure. The proposed method is performed on a number of artificial construction images to show reasonable results.