首先从理论上分析无须重新初始化的水平集方法的主动轮廓图像分割模型，该模型对一些具有不光滑尖角的图像进行分割时，捕捉这些尖角往往不精确甚至失败；然后对利用边缘检测函数的曲率信息识别出凸尖角并进行分割的方法进行研究，由于此方法没有考虑凹尖角的情形，故对含凹尖角的图像分割效果不理想。为解决该问题，提出利用图像的曲率信息识别出凹尖角，再将其与利用边缘检测函数曲率信息识别凸尖角的方法相结合，进一步修正边缘检测函数，达到准确捕捉物体的凸尖角和凹尖角的目的，保证了分割的准确性。数值实验表明，该方法的分割效果较好。

Firstly, an image segmentation method based on level set evolution without re initialization is studied. This method depends on edge indicator function. However, the method is inaccurate in capturing the sharp corners of the object during image segmention. Then, we analyze the method for image segmentation by using the edge indicator function's curvature information to identify the sharp convex corners. However, the model can only capture the sharp convex corners of the object. When the object has sharp concave corners, the model fails to capture the ones accurately. To overcome this difficulty, we employ the image's curvature information to identify the sharp concave corners, and then combine it with the method of identifying the sharp convex corners by using the edge indicator function's curvature information. By further modifying the edge indicator function to make the method capture both the sharp convex corners and the sharp concave corners of the object more accurately. The numerical experiments show the advantage of the improved model.