中文摘要：

提出一种基于过分割的多目标阈值图像分割算法。使用分水岭算法获得待分割图像的过分割区域和分割边界,将类间方差函数和熵函数作为优化目标函数,采用多目标阈值算法对区域的代表点及分割边界上的像素进行划分,再将区域代表点的划分结果扩展到各区域中,以获得整幅图像的分割结果。在多幅Berkeley图像上进行分割测试,并以分割准确率作为算法性能的评价指标,结果显示,新方法在大多数情况下能够获得高于最大类间方差法和最大熵法的分割准确率,此外,由于图像区域信息的使用,使得图像目标能够较为完整地从背景中分离出来。

英文摘要：

A multi-objective thresholding image segmentation algorithm is presented in this paper based on over-segmentation. The watershed algorithm is firstly used to obtain the over- segmentation regions and boundaries of an image; then the multi-objective thresholding algorithm, which adopts the variance between clusters and the entropy as the objective functions, is used to partition the representative points of the segmentation regions and the pixels on the boundaries; finally the partitioning results of the representative points are extended to their respective segmentation regions to acquire the segmentation result of the whole image. Some Berkeley images are adopted in the segmentation experiment and the segmentation accuracy is used as the evaluation index of algorithm performance. Experimental results show that the new method can obtain higher segmentation accuracy than the maximum variance between clusters method and the maximum entropy method in most cases. Moreover, the object can be more completely partitioned from the background due to utilizing the image region information.