Novel methods of 2D linear transform for Face Recognition

• Main Authors: Chih-Ping Chang, 張志平
• Other Authors: Bin-Yih Liao
• Format: Others
• Language: zh-TW
• Published: 2008
• Online Access: http://ndltd.ncl.edu.tw/handle/72758645645111066608

碩士 === 國立高雄應用科技大學 === 電子與資訊工程研究所碩士班 === 96 === In this paper, two novel methods of 2D linear transform for face recognition are presented in order to improve the efficiency of face recognition. The first method, 2DPCA combining with ICA, is proposed to improve the quality of projection conducted by ICA and thus the performance of recognition is improved. Besides, the time-consuming calculation of ICA can be avoided by using the efficiency of 2D concept. The second method, Dia-2DPCA+ICA, is using Dia-2DPCA as the pre-process for ICA so that the performance can be improved. To improve the efficiency of face recognition, eigenvector is requested to be calculated by the method of ICA. If the efficiency of subspace is amended, the quality can be better. According to the theory mentioned, with three common databases in face recognition such as ORL, Yale and AR, the result from the experiments have approved that the algorithm pointed out in this paper have improved the efficiently of face recognition.