Study On The Method Of Image Recognition Based On Fractal Geometry

Pattern recognition of fractographs is very important for analysis of fracture surface.As the development of fractal using in metal fractographs, it has been proved that thefracture surface has the fractal character and can be described with different fractaldimension correctly. Using two methods: vertical section method and box dimension,calculate fractal dimensions regarded as characters of the typical fracture surfaces, such asdimple, cleavage, fatigue and intergranular cracking, the characters are weightedaccording to their own disperse degrees depend on variance. Then design linear minordistance classifier and Bayesian classifier under minimum risk. After training images, thesystem could recognize and classify the fracture surface image. The preprocessing isutilized before feature extraction. The experiment on 100 images of the four fracturesurfaces has a good result. It shows that the calculated dimensions can indicate the fracturesurfaces efficiency, and the two classifiers are useful. This system was completed withMATLAB.