Behavior Analysis of Fractal Features for Texture Description in Digital Images: An Experimental Study

Our Goal • Found that Fractal Features effectively have texture recognition ability using statistical, soft computing, data mining and machine learning methods. • Six texture descriptors introduced in [Chaudhuri and Sarkar, 95] are based on the Fractal Dimension of the original and derived images. The studied features were the FDs of: • the original image (F_1), • the high gray-valued image (F_2), • the low gray-valued image (F_3), • the horizontally smoothed image (F_4), • the vertically smoothed image (F_5) and • the multi-fractal dimension of the original image (F_6). The success of the DBC method is attributed to the fact that differential box counting, between minimum and maximum gray level and the size of the image, gives a better approximation to the boxes intersecting the image intensity surface relative to other box counting methods. Experimental image composed by 9 textures (300 x 300 pixels size) From the 90000 pixels contained in the mosaic, only 41616 corresponds to textures entirely covered by the 33 X 33 operator.