New Unsupervised Image Segmentation via Marker-Based Watershed
一种基于标记的分水岭图像分割新算法

This paper suggests an improved marker-based watershed image-segmentation method to reduce the over-segmentation of the watershed algorithm. The new method applies the watershed transform directly on the original gradients image instead of simplified image, so that the loss of boundary information can be avoided. On the other hand, we design a new marker-extracted approach to extract the regional minima related to the objects from the low frequency components of the gradients. These extracted minima constitute the binary marker image. And then the extracted markers are imposed on the original gradients as its minima, while all its intrinsic minima are suppressed. Finally, the watershed algorithm is applied to the modified gradients by the markers to reduce effectively the over-segmentation. Across a variety of image types, it is proven that this new method can obtain meaningful and homogeneous regions with accurate, consecutive and one-pixel wide boundary. Compared with other methods, this system requires fewer computations and simpler parameters and can more efficiently reduce the over-segmentation of the watershed algorithm.