作者: Xian-Hua Han , Yen-Wei Chen , Xiang Ruan
DOI: 10.1109/ICIP.2011.6116496
关键词:
摘要: In this paper, we propose to use local feature set for image representation, which can represent variations in an object's appearance due changing viewpoint or camera pose. It was evidenced that usually only a part of the object are appeared common when taking photo different view points. With comparison features extracted from positions images, be recognized is two take photos one Canonical Correlation (also known as principle canonical angles), thought angles between d-dimensional subspace, similarity measure sets. The proposed approach evaluated various view-based datasets (Coil-100 and ETH80) category recognition. Experiments show performance advantages our achieved over existing techniques.