分水岭变换在岩屑图像分割中的应用

Detritus image segmentation using watershed transformation

分水岭变换是一种基于区域和数学形态学的图像分割方法,被广泛用于灰度图像的分割之中。但传统分水岭变换过分割问题严重,图像的噪声和虚假纹理会淹没真正想得到的边缘信息。针对岩屑图像的特征,提出了一种改进的分水岭算法分割方案。先在预处理期用形态学开闭重建运算对原始图像平滑处理,在相对保留边缘不受影响的同时,降低噪声的影响。再通过非线性的阈值变换分离出目标和背景,然后在提取出目标的情况下合并过小区域,得到目标的边缘。而由于阈值变换后,区域数量已经明显减少,可以降低区域合并的运算量,提高合并速度。在求取形态学梯度时,选用了一种新的形态梯度形式,消除了形态学处理对分割结果造成的轮廓偏移现象。从实验结果看来,该算法取得了较好的分割效果。

Watershed is an image segmentation method based on region and mathematical morphology, and it is widely used in the segmentation of gray scale image. But the traditional watershed transformation has a serious problem of over segmentation, and noise of the image and fake textures may submerge the wanted edge information. Considering the characteristics of detritus image, this article proposed an improved segmentation method based on watershed transformation. During the pretreatment period, smooth the original image by applying morphological open and close reconstruction operation to reduce the side effects of noises, and relatively remain the edge of the image, then use nonlinear threshold transformation operation to part the background and the targets; after having extracted the targets, merge the small regions, and get the edge of the targets. Because the number of regions has greatly decreased as a result of the threshold transformation, computational complexity of the region merging was reduced. A new form of gradient was chosen to eliminate the drift of contours because of the morphology operation. Judging from the result of the experiment, this algorithm achieves pretty good segmentation outcome.