Research For Fatigue Fracture Image Segmentation Based On BEMD

Metal fracture is one of the three failure modes, and is the biggest danger infailure mode. In metal fracture, fatigue fracture is the main form, it occupied morethan80%of the whole fracture incident. In recent years, metal fracture researchfocused on identification、classification and three-dimensional reconstruction of metalfracture images. However, research on fatigue fracture is limited to the identification,research on texture characteristics of the fatigue fracture images are not much, on thefatigue fracture image texture segmentation are even less. Fatigue striation is thespecific fracture characteristics of fatigue fracture, and is an important basis forfurther research of the fatigue fracture images.Fatigue fracture images of texturesegmentation becomes extremely important. Therefore, the analysis of fatigue fractureis important to determine the failure modes, causes, mechanisms.In image processing, frequency domain analysis in the image feature extractionplays an important role. It is an effective method for texture analyzing of fatiguefracture image. Classical frequency domain analysis methods include Fouriertransform, Gabor transform and wavelet transform. Fourier transform establish thesignal from time domain to frequency domain, but the two are absolutely separate;Gabor transform uses a fixed short window function, which also exists impassabledefects. Wavelet transform depends on the pre-defined filters or wavelet functions.For the drawbacks of the above signal analysis, Norden Huang, etc. in1998proposed a new method of signal analysis and processing-EMD (Empirical ModeDecomposition). EMD is adaptive, and applicable to nonlinear signal andnon-stationary signal analysis and process. Dimensional empirical modedecomposition (BEMD) is an extension of EMD, BEMD is a new hotspot of imageprocessing field.The main contents are as follows:(1)For image preprocessing on the fatigue fracture, there are filtering, based onthe Otsu binarization processing, histogram equalization and so on. Afterpreprocessing, the image can have better image quality and in strong contrast on thegray.(2)Because fatigue fracture image is a kind of complex texture image, it is mainly analysis Fourier transform and wavelet transform texture features of fatiguefracture image in frequency domain. It obtained the texture region of fatigue fractureimage.(3)Using BEMD method to analysis the image texture, segment the synthetictexture image to obtain experiments results. Using BEMD method to filter the imagewhich with a Gaussian white noise, it achieves an ideal filtering effect. Combinationof Laws texture energy description method to describe the texture characteristics offatigue fracture. Using nearest neighbor criterion cluster the feature vectors, andsegment texture region.(4)Finally, compare the effects of BEMD method segmentation results to Fouriertransform and wavelet transform method. BEMD method is significantly better thanthe latter two.This paper mainly used BEMD method to analysis fatigue fracture image. Thismethod can be very good achieved fatigue fracture image texture segmentation.