Contourlet Texture Retrieval Filter Characters

摘要

Contourlet transform has been proved superior to wavelet in representation piecewise texture and contour characters in digital images. In order to improve the retrieval rate further,we studied the representation abilities of some different filters for texture features in two contourlet retrieval systems. The two systems both used Canberra distance for similarity measure,but different feature vectors which composed of energy combined kurtosis and energy combined standard deviation,respectively. Experiment on a 109 texture image datahase from MIT rsults that the filter "pkva" performs better than other ones including bi-orthogonal 9-7,haar,burt,bi-orthogonal 5-3,etc,for the two retrieval systems,under the same system structure.