An affine invariant discriminate analysis with canonical correlation analysis

作者：Lan, Rushi; Yang, Jianwei^*; Jiang, Yong; Song, Zhan; Tang, Yuan Yan

来源：Neurocomputing, 2012, 86: 184-192.

DOI：10.1016/j.neucom.2012.01.026

• Affine invariant discriminate analysis (AIDA)
• Canonical correlation analysis (CCA)
• Central projection transform (CPT)
• Affine transformation

Canonical correlation analysis (CCA) is invariant with regard to affine transformation, but it cannot be directly applied to affine invariant pattern recognition. The reason mainly lies in that many existing CCA-based schemes represent the pattern by matrix-to-vector method, as a result, the structure and spatial information of the original pattern is discarded. In this paper, an affine invariant discriminate analysis (AIDA) method is developed for pattern recognition. Dislike the matrix-to-vector representation, an object is first converted to a projection matrix by central projection transform (CPT). After a point matching process, CCA is performed to projection matrices of the object and the model, and two vectors will be derived. Therefore, the object is classified to a model by the smallest distance between the obtained vectors. Comparisons of experimental results are given with respect to some existing methods, which demonstrate the effectiveness of the proposed AIDA method.

• 出版日期2012-6-1
• 单位中国科学院深圳先进技术研究院; 香港中文大学; 重庆大学; 南京信息工程大学