Application Of Wavelet And Morphology In Image Segmentation

Image Segmentation is the key technology of Image processing, which divide an Im-age into a set of areas with different characteristics and Non-overlapping.The result of image segmentation has a direct impact on the subsequent image processing,Classic image segmentation methods,such as Threshold segmentation can be good at specific images,but no method could be fit for all images.In recent years,with the new technology and theory widely be used to image segmentation,many new methods for image segmentation has been proposed. The improving of existing image segmentation technology has three features,the one is to improve the accuracy of image segmentation result,the another one is to improve the efficency of image segmentation,the last one is to make the method could have a widely application.Watershed Transform is a kind of mathematical morphology segmentation method, through the divide line acquired, the algorithm can get continuous, closed and single pixel wide edge.but it’s shortcoming is very obvious,watershed algorithm is very sensitive to noise and texture subtle change.The faint noise and the sbutle change of texture will lead gradient change greatly,watershed transform implementation on it will cause the serous over-segmentation phenomena,which produce a large number of small areas.To overcome this problem,we improve the preprocess of watershed algorithm,the innovation points as follows:(1) improve the crossover and mutation formulations of the adaptive genetic algorithm and introduce the optimal preservation strategy to improve the algorithm conver-gence.(2) present a new wavelet threshold selection method combine of the characteristics of the image.(3) improve the image denoise algorithm,use the multi-resolution analysis method to enhance the adaptability of denoise algorithm.(4) improve the preprocesion of watershed algorithm to reduce the over-segmentation of watershed.the main work as:(1) perform multi-scale transform on image and take different denoise methods for different subbands to reduce image noise and oversegmentation.As an Image been wavelet transformed,each wavelet coefficient has own characteristics.The low frequency subband with little noise can use the anisotropic diffusion filter to denoise.For the high frequency suband,which contains high noise and details of image,can use threshold atropy denoising.(2) We present a threshold relate to it’s own features and use it to image denoise.This threshold is based on Adaptive genetic algorithm and more close to the optimal threshold than other threshold.(3) Modify adaptive corssover and mutation formulas,and bring the sigmoid function in neural network to smooth the Adaptive crossover and mutation curve to accelerate the algorithm convergence speed and prevent the algorithm from local optimun.(4) According to the marker morphology watershed algorithm to make minimax mark on gradient image,we enhance the edge of gradient image to imporve the accuracy of mark.