Image analysis based on fractional Brownian motion dimension

Yancong Yang,Ruidong Peng

DOI: https://doi.org/10.1109/CSAE.2011.5952414

2011-01-01

Abstract:How to characterize and segment the different textures in an image is a key problem of image processing. All kinds of approaches have been proposed, among which some innovative methods were suggested base on fractal theory. The gray value of images can be mapped to a surface which could be characterized usually by fractional Brownian motion, and thus the fractal dimension were adopted to denote the roughness of different textures and their self-similarity quantitatively. The process of computing the fractal dimension of textures in an image was firstly discussed based on FBM models and then a novel algorithm was suggested to improve the efficiency of calculation, in which only those pixels that lying on the right and bottom edges of the square surrounding window would be considered. When sliding a window with a certain size on an image, the fractal dimension of the window can be calculated and applied to denote the fractal dimension of the center pixel of the window. Thus a fractal dimension image could be generated by mapping the fractal dimension of each pixel. The transformation from an image to its fractal dimension image bring forward a new beginning for further image process and has been investigated and validated through analyzing some SEM images. It's demonstrated that the fractal dimension of textures is little affected by the brightness and contrast of images but determined mostly by the structure of textures. Consequently fractal dimensions could characterize different textures more stably.