一种基于边缘梯度的图像分割方法

利用Krisch边缘方向算子，实现了对彩色和服图片中的和服边缘检测，同时根据真实边界点处的梯度强度值大于其左右邻域的梯度强度值的特征，获得了边缘细化。在结合边界跟踪技术的基础上，有效地去除了和服中大量的与和服分割无关的花瓣图案，得到了和服的有效边缘。实用的边缘连接技术则成功地实现了和服图片中和服各部分的分割。实验结果表明该方法在和服分割上是一种实用的和成功的方法。

An Approach of Grad-based Image Segmentation

In this paper, we implement edge detection of color KIMONO based on Krisch edge operator, and according to the characteristic of which the grad value is bigger in real edge than in its adjoined points, we acquire edge thinning.On the base of contour tracing, we effectively wipe off a great deal of flower pattern unrelated with KIMONO image segmentation, and get a real edge of KIMONO.Then utilizing edge connection techniques, we successfully segment every part of a KIMONO.The experiments show the approach is a practical and successful method in KIMONO image segmentation.