X-ray Image Segmentation Based On Watershed Transformation And Genetic Optimization

As an important tool for clinical medicine diagnosis, the X-ray has a heavy amount of information. X-ray image segmentation is to divide the image into several non-overlap, meaningful and connatural segment. It is not only an important foundation for a doctor to make a correct diagnosis, but also an important basis for further parameter extraction, defected area automatic identification and automatic diagnosis.According to the feature of the X-ray image, this article adopt the watershed transformation method, and make improvement to the way of background marking based on the method of marked watershed segmentation, then a method of watershed segmentation based on double threshold is proposed and has been proven to be feasible to get accurate foreground and background mark. Because the characteristic of X-ray images, it can not obtain accurate segmentation results and lead to an over-segmentation phenomenon even if it is marked accurately for some X-ray image. Wavelet transform is introduced into this article because of the mentioned problem above, and a double threshold watershed segmentation method based on wavelet transform is proposed, that is to proceed wavelet decomposition firstly, and then to proceed double threshold watershed segmentation secondly and do the reconstruction of wavelet in the last. By this way, an accurate segmentation results can be obtained and the over-segmentation is effectively overcome.Structure elements are wildly used in the process of watershed segmentation, and they are chosen by the researchers through continuous experiments, so genetic algorithm is introduced into this article, and the watershed segmentation based on genetic optimization is proposed, and the automatic selection of structure elements is realized. By using this method, a sample image is needed firstly, and by proceeding the sample image the genetic algorithm will be constantly convergent, thereby, the optimal structure element will be obtained, which will be used in the dividing process of similar images. The accurate segmentation results are obtained by simulation experiments, and the results prove the structure elements obtained by genetic optimization are applicable to the dividing process of similar images.