Wavelet Transform Based On Bending Fracture Of Air Component Of The Image Recognition Method

This thesis is supported by the non-destructive testing technology(50775208,51075372), Ministry of Education Research (No.ZD200829003), the traditional image processing method of metal fracture weaknesses, Bandelet transform is applied to the fracture image denoising, enhancement, pattern recognition and neural networks will be introduced to Bandelet transform based identification method of nonlinear feature extraction were obtained better innovation results. The primary contents of this study involve the following:Chapter one discusses the issues raised and their significance. Recognition of metal fracture are reviewed research status and development at home and abroad Bandelet transform the present situation of the main contents of this paper and innovation.Chapter two discusses the wavelet transform theory and image recognition in the fracture shortcomings. On this basis, the article discusses the second generation Bandelet transform the theory and algorithms. The simulation results, indicating Bandelet transform and wavelet transform in image processing in the different fracture and benefits. This chapter is the theoretical basis of the whole thesis.Chapter three introduces the principles of image denoising and methods. In the proposed method, first used with the noise of metal fracture quadtree image segmentation, and get the best of the partition geometry within the flow direction, and the best geometric blocks along the Bandelet direction curvelet transform, the last use of SUREShink calculated Bandelet block adaptive threshold, and then multi-layer soft threshold to remove noise, the Bandelet inverse transform of the reconstructed image. Experiments show that the method is effective, and the traditional wavelet sub-band multi-threshold denoising method compared to the algorithm not only improves the PSNR of the denoised image (PSNR), and have a stronger edge retention.Chapter four, bandelet transform and neural networks combined with the characteristics of each, a transformation based on the metal fracture surface Bandelet nonlinear identification method, the proposed method, using Bandelet transform to extract images of metal fracture Bandelet entropy as the feature vectors, neural network As a non-linear classifier, several typical images of metal fracture experimental verification. Meanwhile, the proposed method with the traditional wavelet transform image recognition method of metal fracture were compared with experimental results show that the Bandelet wavelet transform to overcome the fracture edge in dealing with metal shortcomings, has been compared with the traditional wavelet transform better recognition effect.Chapter five, pulse coupled neural network with (PCNN) and Bandelet transform theory, a new metal fracture pattern recognition method, the proposed method, the fracture after the first PCNN image segmentation, and then transform fracture Bandelet Bandelet image entropy, and will identify as the characteristics of fracture, according to the different characteristics of the image vector Euclidean distance between the numerical size of the MSE identification of metal fracture, experimental results show that:with the BP neural network and Kohonen neural network and other than, PCNN need to learn or training, effective from extraction of complex background signals, coupled with a simultaneous global release and other features. Therefore, the identification of such recognition efficiency is greatly increased.Chapter six sums up the research results of this paper, and proposes the problems which are worth further studying.