This paper presents a through investigation into the use of Gabor filters for the extraction of rotation invariant texture features. Numerous experiments have been conducted to discover the effect of different parameter settings on classification results. The optimum parameter settings are estabilished and tested by classification and content based image retrieval expeirments on a large database of randomly rotated Brodtz texture images. Resistance of the method to Gaussian noise is also examined. The issues studied in this paper are of great imprtance for practical applications but have not been adequately addressed by existing work on texture analysis.
展开▼