返回Google圖書搜尋

Nearest Feature Rules and Dissimilarity Representations for Face Recognition Problems

Mauricio Orozco AlzateGerman Castellanos Dominguez

出版	INTECH Open Access Publisher, 2007
ISBN	39026130339783902613035
URL	http://books.google.com.hk/books?id=aTL6oAEACAAJ&hl=&source=gbs_api

註釋In this chapter, we presented a series of theoretical and experimental considerations regarding the nearest feature rules and dissimilarity representations for face recognition problems, analyzed separately as well as a combined approach. Firstly, a study about the asymptotic behavior of the nearest feature classifiers was conducted, following the wellknown procedure derived for the k-NN rule. We concluded that, if an arbitrarily large number of samples is available, there is no significant difference between k-NN and its geometric generalizations: the nearest feature rules. Moreover, as for k-NN, it is not possible to say something general about the asymptotic behavior in the finite-sample case. It might be possible to perform an analysis for specific distributions; perhaps without loss of generality. Consequently, further conceptual considerations and experiments are required. Quantifying the computational complexity of classifiers is very important in the selection of a particular algorithm. Complexity of algorithms is usually measured in terms of orders; nonetheless, such an approach is not precise. An evaluation of the error-complexity trade-off for the nearest feature classifiers has been presented in Section 3. We have also studied the complexity of nearest feature classifiers, in terms of the number of additions and multiplications associated to their evaluation, as well as through error-complexity curves and a comparative study considering error and complexity. It was shown that k-NFP is too.