The Research Of Center Location Algorithm Of Many Quasi-circle Objects

With the rapid development of computer technology and machine vision, digital imageprocessing technology has been used widely in target statistics application. Automatic countingfor quasi-circle objects becomes one hotspot of present image processing research. It is possibleto reduce the labor-consuming, improve the efficiency and accuracy to count the number ofobjects automatic by computer technology.A center location algorithm for quasi-circle objects is proposed for bar images, whichpurpose is to count the number of objects according to the detected center points. At first, thispaper analysis the characteristic of bar images in detail. With the trait that the edges of objectsare symmetrical in all directions while the edges of background are not regular, this paperproposes a method for extracting region of interest based on directional filtering and directiondetection. After that, this paper also describes image pre-processing technology which are usedin this center location system.Secondly, This paper summarizes and analysis several common image adhesionsegmentation methods used for overlapping quasi-circle objects. According to the actualcharacteristics of the quasi-circle images, this paper proposed a target segmentation and centerlocation algorithm for bar images based on Hough transform. In this method, the gradientinformation of object edge points is obtained firstly using edge detection operator, then improvesthe Hough transform algorithm. This method works well for the segmentation of overlappingquasi-circle objects.In the process of center decision, this paper designs a rule for center decision, Because thegray value of the central region are greater than other regions, and there is one center point in therange of object’s radius. The algorithm of center decision is simple to realize and works well toremove the false points.At last, with the center location algorithm for quasi-circle objects proposed in this paper,design image processing software base on Visual Studio2010and OpenCV, and take amount ofbar images for experiment. Experiment results show that the algorithm locates the center pointaccurately, and improves the counting accurate rate significantly than the improved houghtransform algorithm based on annulus.