Texture classification using Retina Cell- based Model Filter Bank

Feature extraction and similarity measurement are lhe two common stages used rn classifcation purposes In order to improve the total classification quality in texture analysis, a new ftlter bank design is introdttced. This new spatial-frequency quantization i,s based on non-uniform sampling principle of retina cells. This approach provides capturing more detail informalion arottnd the origin of Fourier region and less information farther from lhat point. A cone beam form sampling strateg) has been introduced, where the size of each cell increases proportionally to the inverse of the cell's distance from the DC region The used feature vector is Modifed Absolute Average Deviation (MAAD) from mean. The feature extracted is classified implying maximum likelihood classifier with equal number of 288 samples used for training and testing phase Experimental results on I6 Brodatz texture images indicate that the new method signijicantly improve the classification rate; e.g. from 89% to over 93?6 compared with Gctbor feature, as a well-known method in texture segmentdtion

Keywords:

Texture analysis , feature extraction , modelling and simulation , non-uniform spatialfrequency domain quantization.