抄録
This paper describes a method to separate and count the number of particles in an overlapped image. The edges in an image are detected by Canny's method, and contours are linked from the pixels in edges. Contours are segmented using a criterion based on the derivative of the curvature. Constant curvature segments are clustered according to the relations among contour segments which are likely to represent the same circles. A minimization criterion is used to find out the circle parameters that provide accurate parameter estimation. The method can recover many circles in a heavily overlapped particle image, and provide a useful way in some practical applications, such as counting particles, separating overlapped bubbles in PIV system.
