一种新的边缘保持分水岭的图像分割算法

A Novel Segmentation Algorithm Based on Edge-Preserving Watershed

为了达到抑制分水岭过分割和保持物体边缘信息不受破坏的双重目的,提出了一种新的边缘保持分水岭(Watershed)算法。首先,根据K-均值聚类将图像分成多块;然后利用噪声标准差构造相对应的双边滤波器平滑每块图像;接着计算形态学梯度,对梯度图像进行H-minima标记;最后对标记图像进行分水岭分割。该算法将双边滤波和分水岭算法相结合,有效地抑制了过分割并且较好得保持了物体边缘信息。

A new image segmentation algorithm based on edge-preserving watershed is presented to solve the conflicts between suppressing the over-segmentation of watershed and preserving the object edge. First, the noise level of each segment is estimated after K-means clustering, and then the bilateral filter is applied by using the noise level. Second, the morphological gradient is calculated and the marker extraction of H-minima is used. Finally, the watershed algorithm is applied to the gradients. Compared with other methods, over-segmentation can be significantly reduced while boundaries of major objects are precisely located by incorporating bilateral filtering with watershed transform in this system.