Segmentation of dense grain images based on level set model

In experimental soil mechanics, it is important to study the motion of sand grains under high pressures. Before measuring the movement and deformation of a fraction of sand, we should detect it at first. But it is very difficult to distinguish sands automatically from each other in a dense grain image. In this paper, an approach of segmentation based on level set is proposed for dense sand grain images. The appropriate initial position is essential for level set method. Firstly, we segment the sands by boundary exploration method. The gained contour outlines roughly the sand and is used as the initial position. And then level set is employed to obtain the exact shape. In order to improve the speed and avoid overshooting, an ending condition is designed. Several experiments are performed. The promising segmentation results can be obtained on dense grain images with this approach.