Processing Method Of Metal Fracture Images Based On Contourlet Transform

This thesis is supported by the National Natural Science Foundation of China (No.51075372) and the open Fund of Key Laboratory of the Ministry of Education of Nondestructive Testing technology (No.ZD200829003), introduces Contourlet transform to the metal fracture image processing, such as image denoising, enhancement, fusion and recognition. Some preferable innovative achievements are obtained in this paper. The primary contents of this article involve the followings:Chapter one illuminates the significance of proposing and studying on this thesis, summarizes the research status at home and abroad of the metal fracture image processing and Contourlet transform, and presentes the primary contents and innovation points of this paper at last.Chapter two introduces the theory of the Contourlet transform due to the shortages of wavelet transform in image processing area. At the same time, the advantages and disadvantages of the Contourlet transform and the wavelet tranform are compared, and the characteristics of the pyramidal directional filter bank Contourlet are given. The content of this chapter is the basic theory of the whole thesis.Chapter three introduces an algorithm based on non-redundancy Grouplet-Contourlet transform in image denoising and enhancement as the low-frequency decomposition of the Contourlet transform has a redundancy up to33%and the new transform Grouplet proposed by Mallat in2008, which is implemented with an orthogonal weighted Haar transform that adapts the lifting parameters to groupings specified by multiscale association fields, can exploit the geometry of the image. At the same time, the proposed method is compared with the methods based on wavelet transform and Contourlet transform in image denoising and enhancement. The experimental results show that the proposed method is superior to other methods. It not only improves the PSNR of the denoised and enhanced image, but also retains good contour information of the image.Chapter four combining a non-redundancy algorithm based on WBCT—a new non-redundant and perfect reconstruction transform, which can exploit more information of the image’s direction and texture and the latest research results of the Biology and Neuroscience in recent years—PCNN, which can extract useful information from complex background and does not require learning or training, proposes a fusion algorithm based on WBCT-PCNN. In the proposed method, firstly, two original images were decomposed by using WBCT. Then, the low frequency subband coefficients and a series of bandpass directional subband coefficients were obtained. All of fused subband coefficients were determined by image fusion algorithm based on PCNN. Finally, by performing the inverse WBCT on the fused subband coefficients, the fused image was obtained. The experimental results show that the fusion method is superior to fusion methods based on PCNN, wavelet-PCNN, Contourlet-PCNN and NSCT-PCNN.Chapter five introduces kurtosis into the field of image processing and defines the concept of Contourlet kurtosis. Kurtosis is more sensitive to the texture of the metal fracture than L1norm and the average energy, while it is not sensitive to the direction of the image. On this basis, a recognition method of metal fracture image based on Contourlet kurtosis is proposed. In the proposed method, three-level Contourlet transform is applied to each metal fracture image firstly, and the kurtosis of each Contourlet transform frequency band output is used as the characteristics of fracture image recognition. Then K-nearest neighbor classifier is used to classify the fracture images. The experimental results show that the proposed method is superior to the recognition method based on L1norm and the average energy calculated by Contourlet coefficient.The relevance vector machine has a good generalization ability and can give a probability measure on the ownership of the categories. Combining the respective advantage of Contourlet transform and RVM, a recognition method of metal fracture image of aerial material based on Contourlet-RVM is also proposed in this chapter. In the proposed method, two-lever wavelet transform is applied to metal fracture image recognition by selecting Contourlet transform. L1norm of each wavelet transform frequency band output is used as the characteristics of fracture image recognition. And RVM classifier is used to classify the fracture image. At the same time, the proposed method is compared with the Contourlet-SVM recognition method. The experiment result shows that the proposed method is very effective. Whether in the correct recognition rate, or in the training speed, the Contourlet-RVM recognition method is superior to the Contourlet-SVM recognition method.