一種基于奇異值分解的特征抽取方法
A Method of Feature Extraction Based on SVD
-
摘要: 特征抽取是模式識別的基本問題之一,Fisher線性鑒別分析是特征抽取中最為經(jīng)典和廣泛使用的方法之一。該文分析了Fisher線性鑒別分析在求解過程中可能存在的問題:鑒別矢量的分量可能是復(fù)數(shù);特征值對擾動的敏感性;鑒別矢量之間未必具有正交性。由此提出了均衡散布矩陣的概念,并利用均衡散布矩陣構(gòu)造了一種新的線性鑒別準則。利用奇異值分解定理,將求取鑒別矢量轉(zhuǎn)化為對矩陣求奇異向量。用該方法進行求解可以有效地避免前述的問題。試驗結(jié)果表明,該鑒別準則具有良好的鑒別能力。
-
關(guān)鍵詞:
- 模式識別; 特征抽取; 線性鑒別分析; 奇異值分解; 人臉識別
Abstract: Feature extraction is primary problem of pattern recognition. As one of the most classic methods in the field of feature extraction, Fisher linear discriminant analysis is applied widely. It may meet several possible problems in finding optimal set of discriminant vectors: the components of these vectors may not be real; the eigenvalue may be sensitive; these vectors may not be orthogonal each other. So the balanced scatter matrix is proposed in this paper. Based on the matrix, a discriminant criterion is formed. The optimal set of discriminant vectors can be acquired througn singular value decomposition theorem. The method can avoid the problems metioned above. The result of face recognition experiment shows that it has powerful ability of feature extraction. -
Fisher R A. The use of multiple measurements in taxonomic problems. Ann. Eugenics, 1936, 7:179- 188.[2]Wilks S S. Mathematical Statistics. New York: Wiley, 1962:577 - 578.[3]Duda R O.[J].Hart P E, Stork D G. Pattern Classification, 2nd Edition. New York: John Wiley Sons.2001,:-[4]Fisher R A. The Statistical Utilization of Multiple Measurements.Ann. Eugenics, 1938, 8:376 - 386.[5]GolubGH,Van Loan C F.袁亞湘等譯.矩陣計算.北京:科學(xué)出版社,2001.[6]Belhumeur P N, Hespanha J P, Kriegman D J. Eigenfaces vs.Fisherfaces - recognition using class specific linear projection.IEEE Trans. on Pattern Anal. Machine Intell, 1997, 19(7): 711-720.[7]Foley D H, Sammon J W Jr. An optimal set of discriminant vectors[J].IEEE Trans. on Computer.1975, 24(3):281-[8]程云鵬.矩陣論.西安:西北工業(yè)大學(xué)出版社,1989:294-302.[9]Martinez A M, Kak A C. PCA versus LDA[J].IEEE Trans. on Pattern Anal. Machine Intell.2001, 23 (2):228- -
計量
- 文章訪問數(shù): 2494
- HTML全文瀏覽量: 86
- PDF下載量: 853
- 被引次數(shù): 0
下載:
