Image-based Bottom-up segmentation method can acquire the edge of the object with quite high exactness according to the discontinuity of the image, while the Top-down segmentation algorithm can acquires the configuration expression of the object, I. E.' CSF (class-specific fragment) through learning, where the CSF can be used to cover the images to be segmented and helps to obtain better segmentation effect. By combining the advantages of both criteria, we present an effective segmentation method. In the approach, the Top-down segmentation results are taken as the input images of Bottom-up segmentation, and the modified multi-scale marker is employed to control the refined edges of watershed transform to accomplish the segmentation. The algorithm uses the similarity standard and can determine conveniently and simply the sub-block belonging to the CSF and matches and covers the object image. Meanwhile, the morphological reconstruction computation is added into minima imposition technology, after the complementary operation being conducted on its result, the regional minimum is obtained, and the marker model constructed is more ideal. Experiments demonstrate that, the segmentation results achieved by this method are superior to those obtained by Top-down and Bottom-up method alone. Even in these parts where the backgrounds are variable and the bodies are movable, the object can still be well segmented.%基于图像的Bottom-up分割方法能够根据图像的不连续性以较高的正确率获取目标边缘；Top-down分割方法通过学习可以获取目标的外形表示,即对象基元CSF( Class-specific Fragment),该CSF能够用以覆盖待分割图像,帮助获得更好的分割效果.结合这两种准则的优势,提出一种有效的图像分割方法.把Top-down分割结果作为Bottom-up的输入,用改进的多尺度标记控制分水岭变换细化边缘,完成分割.算法利用相似性标准,能够简便快捷地判定属于对象基元的子块并匹配覆盖目标图像；同时,把形态学重构运算加入到强制最小技术中,对其结果求补后获取区域最小值,构建出的标记模型更加理想.实验表明,该方法得到的分割结果优于单独使用Top-down和Bottom-up方法,即使在背景可变和身体可动部位也能较好地分割出目标.
展开▼
