A fast and accurate method for counting and segmentation of granular objects by using image processing

Counting the number of objects using image processing is very important in many aspects from medical science and biology to agronomy, genetic study and plant breeding. Although many studies were implemented for detecting the objects in image processing but the counting objects is easier and faster than recognition, because in some cases, an object can be counted and not be detected. In this study, a new algorithm is presented to count the granular objects without detection and recognition. This method can work correctly in despite of some noise in image. Presented method is based on distance-transform and because of this, the results almost are independent on size and depend on shape of objects and in other hand this method is relatively simple and fast. The other strength of this method is to use from detected edges. Not only perfect edge is not needed but also even one pixel from edge in some cases can be effective for counting correctly. The sharper edges are more helpful in this method. This is a good combination of detecting edge and using shape and size to count correctly. The results showed that in all of conditions, ideal or natural, the proposed method was better than the other investigated methods. Some of investigated methods were implemented only in special condition or special shape of objects but results showed that our method is capable in all of the investigated conditions. Disadvantage of this method is that it does not work on star shape objects unless changes to convex shape with morphological transforms. In spite of the good results, sharp edges and a good binary image of objects guarantee an acceptable counting result. The presented segmented image as graphical output is a valuable instrument for object recognition in machine vision systems. This method is recommended for counting plant seeds, especially convex seeds.