Automatic segmentation and counting of touching bar images

A method for automatic segmentation and counting of touching bar images is developed to solve the problems of bar images automatic counting. Firstly, the granulometry is used to obtain bar radius to adaptively select correct parameters. Then a novel method for computing the center area of the bar is introduced to avoid initial over-segmentation markers and the concept of strong touching, and the weak touching is put forward. A two-step marker method is proposed according to touching relationship. After eroding the image is classified and the initial marker result is obtained. Under segmentation marker areas are eroded step-by-step, and the knowledge of bar shape and area is used to obtain right image markers. Finally, two-step marker results are fused and the image is segmented with a marker-based watershed algorithm. Experiments show that the method can effectively segment touching bar images and obtain reliable the number of bars.