appearance-based two-dimensional models of image variability compared LDA to the maximum entropy approach directly.

compared LDA to the maximum entropy approach directly.

5.4.2 Experimental results In a first experiment, we compared LDA to the maximum entropy approach directly. We observed the error rates on the USPS task using single Gaussians with pooled diagonal covariance matrix, 9-fold virtual training data and an LDA estimated on 40 clusters or pseudo-classes yielding a 39-dimensional feature space.

file:///C:/Users/msfcfname/Downloads/dissertation-keysers--modeling-of-image-variability-for-recognition--2006.pdf