2-D Images and 3-D Objects Normalization Using Moment-based Covariance Matrices

• Main Authors: Chia-Che, Wu, 吳佳澤
• Other Authors: Soo-Chang, Pei
• Format: Others
• Language: en_US
• Published: 2000
• Online Access: http://ndltd.ncl.edu.tw/handle/86547976629459911562

碩士 === 國立臺灣大學 === 電信工程學研究所 === 88 === Abstract Many researchers have proposed invariants under geometric transformations, but they are only invariant under translation, scaling and rotation. These invariants can’t recover the skew transformation. Also, the computation cost of the invariants is so high that recognition becomes slow. In this thesis, we propose a normalization method to recognize objects. After our normalization algorithm, all geometric-distortion objects become unique-forms. Thus every simple feature can be used to distinguish different and same objects. The computation is very fast since the normalization only uses second order moments. The features are also computation-less and can have large amount. We develop a complete algorithm set in 2-D and 3-D object normalization. This algorithm can normalize objects under general geometric transformation, including translation, scaling, rotation and skew. That means we can recover objects under these transformations. Also, the blurred image restoration is researched by us and has a good result.