V.A Trenikhin, V.G. Kobernichenko
The most popular methods of determination fractal characteristic for practical aims of remote-sensed data processing are the Herst index method, the local-dispersing method, the covering method, the method of prism and so on.
This article proposes the modification for one of the most popular algorithms based on the local-dispersing method for determination fractal dimension.
The classical local-dispersing method uses the method of linear window smoothing for getting the set of different scaled images. The proposed modification uses wavelet-transformation for this aim. The dispersion estimation of brightness is realized in sliding window. In the first step, the forward wavelet transformation based on Haara wavelets is used for processing image area defined sliding window. Second step is the inverse wavelet transformation. But detail coefficients are discarded in this step. The result of processing is the sets of dispersion estimation for different scaled images. The first set is defined initial images; subsequent sets are defined degree of wavelet transformation.
The article describes the authors developed computer module for obtain the fractal characteristics and wide estimation received result.
Some experiments were provided for different algorithms comparison. The local-dispersing method, the modification of local-dispersing method and the method of prism ware tested. The initial dates were multispectral, panchromatic and radio-locating images. The fractal dimension field was made in the brightness gray scale from 0 to 255, and 0 corresponds to minimum fractal dimension, 255 – maximum.
In Result, experiment demonstrated scaling and fractal behavior for each initial image. The thresholding of fractal dimension field is the instrument for clusterization and classification poorly distinguishable areas and boundaries. The most important result is that, that fractal dimension has one mean for one-typed areas and different for different areas. This results demonstrate fractal behavior different natural textures