Performance Estimate of Range Profile Feature Extraction for the Case of Defined Viewing Aspect by Means of Fisher Score

In this article performance estimation technique of range profile feature extraction is described. This technique is suitable for the cases when range profiles are acquired with an a priori knowledge of viewing aspect of an object. The feature extraction performance is estimated by separability of feature vector sets provided by the procedure of extraction. It is shown that in case of know viewing aspect, Fisher score could be used as a measure of separability. Comparison of Fourier transform based and wavelet transform based feature extractions is provided. Provided modelling results shows relation between Fisher score-based estimate of separability and AUC ROC classification performance estimate.