基于边缘预测与边缘增长的图像分割方法

Image segmentation based on edge prediction and edge growth

针对现有图像分割方法过分割与分割效率低的问题,该文提出了新的基于边缘的图像分割方法。算法首先在图像梯度基础上提取锚点得到初始边缘图像,从初始边缘任意一点沿两个不同的方向出发,进行边缘预测,遍历边缘图像形成边缘段。然后,通过八邻域扫描的方法寻找边缘断点,根据断点间的马哈顿距离与距离阈值寻找断点间的最小距离并连接。进一步,判断剩余断点与其周围像素点中距离最小的像素点并连接,消除断点。最后通过判断边缘增长的增长点周围的平均灰度值与断点处灰度值的差值是否满足阈值要求,满足即在11×11邻域内进行增长直到遇到其他断点或是边缘,否则在15×15邻域内进行增长。实验结果表明,本算法能够通过边缘信息快速连接获得封闭区域,图像分割速度快、准确性好。

A new method of image segmentation is proposed for the realistic situation of image segmentation and segmentation efficiency. First,the proposed method extracts the anchors to get initial edge image based on the image gradient,and then predicts edge segments by linking the continuous edge points which starts from the initial edge point in two different directions. Second,the breakpoints are found based on 8 neighborhoods scanning and the small gaps between breakpoints are connected based on the smallest Manhattan distance and distance threshold. Finally,the edge segments are confirmed whether growth by judging the difference of the gray value between growth point and its neighborhood breakpoint. If the difference meets the threshold requirements,the edge segment will grow in its 11 × 11 neighborhood until other breakpoints or edges are met,otherwise the edge segment will grow in its 15 ×15 neighborhoods. Experimental results show that the proposed algorithm can quickly connect edge information to obtain closed regions,the segmentation speed is fast,and the accuracy is good.