Vehicle verification using Generalized Gaussian Distribution feature descriptor

A new feature descriptor based on the statistical modeling of wavelet output is proposed for vehicle verification problem. The Generalized Gaussian Distribution (GGD) offers accurate statistical fitting for the real and imaginary part of Gabor and Dual-Tree Complex Wavelet Transform (DT-CWT) output. Consequently, the descriptor is simply composed from the GGD parameters of Gabor and DT-CWT output by means of Maximum Likelihood Estimation (MLE) approach under GGD assumption. Experimental result shows that the proposed feature descriptor outperforms the former existing method in the vehicle verification using SVM classifier.