Rough Set Theory And Genetic Algorithm-based Digital Image Processing Algorithms

Medical image includes lots of pathological information. It is important to clinical diagnosis and treatment. The computer medical image processing has been a hotspot to researchers at home and abroad. Therefore, it is very significant to explore more accurate and higher-speed automatic computer medical image processing.The task of medical image processing includes the image pretreatment, such as denoise and enhancement, segment, image registration and fusion, feature extraction and 3D reconstruction, and so on. The content of this dissertation is the image pretreatment and segmentation. In order to establish the foundation for the next more complex processing, this dissertation attempts to apply several new theories and methods. Medical image processing is the part of digital image processing. So we can process the medical image by general methods. However, the medical images have the characteristics of themselves. The general methods can't gain the ideal result. Therefore, on the base of predecessor's work, some new theories and methods are introduced to improve the general methods in this dissertation.We use wavelet transform to decompose the fuzzy enhanced image by the multi scale function of wavelet, and then extract the edge by the improved porosity algorithm, according to the relationship of the module maxinum value and the edge. After tracking and compensating the source edge, we can get the ideal edge.In the dissertation, we use the genetic algorithm to optimize the parameter of the image processing, such as the parameter optimization of fuzzy enhancement, the segment based on the splitting and merging algorithm and the motif extract, and so on. The results are proved effective.The application of rough sets in this dissertation is a new point: according to the class attribute of rough sets, we can divide the image into the marginal zone and non-marginal zone, and then enhance them separately. From the results, the method is proved effective.At the last part of this dissertation, the multi scale decomposition of wavelet, rough sets classification, fuzzy enhancement and genetic algorithm optimization are combined together to enhance the image. The template having the invariable scale and rotation character and the genetic algorithm optimization are used to segment the enhanced image. By the experiment, this method is proved effective. However, the speed of the method is not fast.The methods in this dissertation are simulated in the environment of Matlab 6.5 and Visual C++ 6.0.