A Feature Extraction Method Based on Correlation Matrix

• Main Authors: Chang, Weiming, 張偉民
• Other Authors: Kuo, Borchen
• Format: Others
• Language: zh-TW
• Published: 2012
• Online Access: http://ndltd.ncl.edu.tw/handle/59531876728620879763

碩士 === 國立臺中教育大學 === 教育測驗統計研究所 === 100 === Recently, the high correlation property between neighboring bands is usually used for dimension reduction on data clustering or classification by grouping similar bands. However, there are two main difficulties. One is how to cluster similar bands based on the correlation matrix of bands. The other one is how to determine the thresholds for splitting the similar bands. Fortunately, the spectral clustering, a clustering algorithm based on a similarity matrix, can be used to solve the two problems simultaneously. In this study, we propose an unsupervised feature extraction method based on the correlation matrix of bands. The spectral clustering based on fuzzy c-means is applied to the correlation matrix of bands (CMFESC), and the corresponding membership values determine the transformation matrix. Experimental results on the educational measurement dataset, Indian Pine Site dataset, Washington DC Mall dataset, and some UCI data sets, show that the proposed method achieves good segmentation performance compared with principal component analysis (PCA) and independent component analysis (ICA).