Rock Fracture Image Processing Based On Wavelet Analysis

It is an important task to detect rock fractures in rock engineering. Such as the radionuclide transport and retention processes in fractured rock are very necessary in nuclear waste management. The techniques of image processing and analysis can be applied as a power tool to obtain more detailed information of rock images.Wavelet analysis is internationally recognized up to minute tool for analyzing time frequency and is widely applied to signal processing. This thesis discusses the technique of image processing based on wavelet transform.There are many methods to obtain the rock fracture images. The inner fractures image can be obtain by using ultraviolet; on the other hand, the external fractures image can be obtain by using visible light. The methods are efficient and low cost. To detect the ultraviolet image fractures, the thesis presents an algorithm based on multi-scale wavelet transform. The method is useful not only to rock fractures detection but also to other images edge detection.The color images are acquired by using visible light; moreover the fractures are more complicated. In order to suppress noise, a novel fracture detection algorithm based on quaternion convolution and wavelet transform was presented in this thesis. Once the color image was convoluted using different scale quaternion operators, the dot product is applied. Finally, the edge map could be obtained by using modulus maxima suppressed. This method can obtain accurate edge location meanwhile, it is efficient for noise suppresses.Because the visible light image is colorful but noisy and the ultraviolet image is edge clear but grey-scales, the ideal solution is fusing the two types of images. After the color image was transformed to IHS color space, the edge information is fused in different areas. The fused images contain the edge of the ultraviolet image and the color of visible light image, which are more useful for further processing.The thesis also presents a motion deblurring algorithm based on estimation. First of all, the blurred image is sampled at blurred direction, and each sub-image is equal at the same location. Depending on the character, the original pixels can be computed and other pixels also can be calculated. The proposed method is evaluated by both simulated data and real images, at the same time the restoration qualities of different deblurring methods are compared, and the experimental results show promising results.