K-L Transformation Based on Cosine Angle Distance

Classical Karhunen-Loeve(K-L) transformation is based on Euclidean distance, and Euclidean distance is sensitive to outlier. In many cases, cosine angle distance has better performance than Euclidean distance. In this paper, a K-L transformation based on cosine angle distance (K-L-C) algorithm is proposed. K-L-C transformation uses cosine angle distance to measure the error of data reconstruction in the process of searching for the optimal represented bases. Experimental results on YALE face database and PolyU palmprint database show the superiority of K-L-C transformation over K-L transformation.