| With the advent of information age, a lot of image information need for real-time and accurate processing, which gives people a great of challenges in the application of imaging technology. Because image segmentation is the foundation of image recognition and understanding, it is a key of the image engineering. But the current image segmentation is lack of generality and poor in preprocessing image noise for image segmentation, so developing a new image segmentation method to solve the current problems is greatly important.In this thesis, firstly, we describe the classical image segmentation methods systematically, including their principles, characteristics, advantages, disadvantages and applicable conditions. Then, we introduce the traditional method of dealing with image noise-filtering, and discuss that the traditional approach has the adverse effects of processing image noise on image segmentation. Using the traditional image in filtering denoising, some important details may be lost, and sometimes the image processing would be desirable. Then, we focus on a new image segmentation method and the process of the realization based on the statistical properties of noise, and gives effective results. The process of the realization includs the initial segmention of original image, region merging of the overlapping point by the statistical model of noise. Therefore, compared to the traditional image segmentation method, the method of this thesis in stability of image segmentation and in excessive segmentation have greatly improved and the noise of statistical model parameters- mean and variance are used to describe merge criteria. By experimental analysis, we can prove that this new image segmentation method has the certain innovation and efficiency. |
