Study On Segmentation And Recognition Algorithms For Underwater Images Based On Fractal Theory

As a math tool that describes the extraordinary abnormality and complexity in nature, fractal theory has been widely used at image processing and image analysis. The thesis is devoted to the applications of fractal theory to the segmentation and recognition for underwater image.At first, the basic theory and development history of fractal are introduced. The author classified two kinds of box fractal dimension and presented the methods to computing them. Then, this paper generalized the basic theory and character of FBM(Fractional Brownian Motion), two methods of computing the Hurst exponent were proposed.Multifractal theory is an effective tool for singularity processing and has turned to be the focus of research on fractal. The calculation of multifractal spectrum is important but difficult, this paper proposed the so-called wavelet transform modulus maxima (WTMM) method to get the multifractal spectrum of underwater images, the results show that the multifractal spectrum effectively characterized three kinds of texture (stone, sea water, hot liquid). a new feature parameter S ( d ) was proposed which described the variations of point-wise; together with the local singularity exponentαand the spectrum f (α), the author presented a new segmentation algorithm for underwater images.We analyzed the increment distribution and the self-similarity of several typical objects of underwater, and the results show that the self-similarity parameter is a variable with respect to measure scale. Combining the self-similarity parameter with two increment parameters to the recognition eigenvector, we studied the separability tolerance of the eigenvector and presented the method for feature selection. The results show that Correct identification rate was 96% which satisfied the underwater robots for automatic identification.