Research On Improved Watershed And FCM Algorithm Applied To Image Segmentation

Image segmentation technology is an important foundation for image analysis and image understanding, and is widely used in industry, medicine, geography, transportation and other fields.It is more and more attention from scholars.The innovation of theory makes the image segmentation algorithm based on special theory emerge in an endless stream.Watershed algorithm based on morphological theory has a high precision and speed, but it has a serious problem of over-segmentation, fuzzy c-means (FCM) algorithm based on fuzzy theory has the advantages of automation, and can solve the problem of multi branch segmentation, but it is greatly influenced by the initial value.According to the needs of the subject application, in order to get a better segmentation result, this paper studies the improvement of these two algorithms.Firstly, the advantages and disadvantages of watershed algorithm are analyzed, the classical algorithm is improved from the preprocessing stage, an image segmentation of marked-watershed based on morphological filter is considered to be adopted. The improved algorithm uses opening-and-closing-by-reconstruction filter to filter the gradient image, and takes advantage of internal and external markers to control the segmentation region, then the gradient image with tags is segmented by watershed algorithm, the method is applied to magnetic resonance imaging (MRI) brain tumor image segmentation and river synthetic aperture radar (SAR) image segmentation, eliminates the over-segmentation phenomenon.The simulation and contrast experiment results show that the method has high segmentation accuracy and speed for magnetic resonance imaging (MRI) brain tumor images, affected by speckle noise and shadow, the synthetic aperture radar (SAR) image segmented by the method is not ideal, at the same time, as the influnce of threshold value of the extended extreme value transformation, the appropriate threshold value is determined artificially according to the simulation test to get the desired result.Secondly, an improved fuzzy c-means (FCM) image segmentation method is studied, in order to overcome the shortcomings of FCM algorithm, which need to determine the number of categories artificially, and easily fall into local extreme value. The improved algorithm takes use of the Kuan filter to filter the original image, gets the initial class center and the number of categories according to the histogram of the filtered image, these data guides algorithm for FCM image segmentation, so this algorithm can adaptively determine the number of categories and the initial class center, compared with the classical FCM algorithm, the image segmentation accuracy is high, and the speed is fast, the method is applied to river SAR image segmentation, detection and recognition, compared with the improved watershed algorithm, image segmentation effect is improved, and the river recognition rate is improved.In summary, there are many classical algorithms and improved methods for image segmentation, the segmentation effect is different, to obtain a better segmentation result for images to be segmented with different attributes, image attribute, target, algorithm complexity, computational efficiency, impact factor, application background and demand and other factors need to be considered overall in application and improvement.