The Research Of Image Retrieval Method Based On Semantics

An image classification method based on phase congruency has been proposed,which concerns the following scenes: Indoor and Outdoor, Building and Landscape, Dayand Night, and the location and recognition of trees without leaves has also achieved.This thesis analyzed the method of image feature detection based on grads;logGaborwavelet is analyzed in the human visual systems characteristics viewpoint;we discusssthe importance of image phase information, utilize the phase congruency to detect imagefeatures, distinguish different features according to local phase information, get thestatistical histogram that describe image features, include step edges, lines and roofedges, Mach bands and then the semantics description of image;when calculating thephase congruency, we estimate the noise level, they are finished in the same time. Theapplication of Support Vector Machines(SVMs) in the image classification isinvestigated, to improve the accuracy, a reject scheme is adopted in our system throughrevising the standard SVM output into the probabilistic output. Finally the summary andprospect of the work is made.