Principal Component Analysis

The result of feature extraction is a 316-dimensional feature vector describing various gray-scale histogram, textural, and morphological characteristics of each region. The curse of dimensionality [27] is a serious constraint in many pattern recognition problems and to maintain classification performance, the dimensionality of the input feature space must be kept to a minimum. This is especially important when using an ANN classifier, to maintain a desired level of generalization [32]. Principal component analysis (PCA) is a technique to map data from a high-dimensional space into a lower one and is used here for such a purpose.

To use PCA in the adaptive knowledge-based model in an unbiased way, the PCA coefficients, comprising eigenvalues and eigenvectors, are determined from an independent training set. In mapping to a lower dimensionality, only eigenvalues > 1.0 are considered and the eigenvectors from training are applied to a testing pattern. Testing and training folds are formed using 10-fold cross validation [32] such that an unbiased PCA transformation can be obtained for each testing sample.

0 0

Post a comment