Algorithm Research And Software Realization On Colony Image Segmentation

Colony image segmentation is a tool for capturing colony objects by separating colony targets from the background via a specified algorithm.So far,it has been applied to different fields including counting of colony objects and classification of colony targets,due to its accuracy and rapidity.The main contents of this dissertation are summarized as follows.First,we introduce some existing colony image segmentation methods including their developments,merits and demerits.The Otsu-based segmentation method,frequently used in segmenting simple colony images,is introduced in detail.Some experiments are also performed to analyze the property of this method.Secondly,in order to overcome the limitation that those exiting segmentation algorithms cannot accurately segment overlapping colony objects with small sizes,we present an algorithm based on ellipse fitting to segment colony images with overlapping objects.The proposed algorithm consists of the following four components.(1)Pre-segmentation and boundary point detection.In this procedure,a method based on Otsu is used to obtain the binary image,followed by a boundary detection algorithm to search the visible edges of colony targets.(2)Seed-point detection.Here,the Distance Transformation(DT)and the H-minima transformation are combined to gain seed points corresponding to each colony object.(3)Edge to seed point association.We use the distance between the boundary point and the seed point,and the intersection angle between the connection of them and the gradient vector of boundary point,to set matching criterion,for the purpose of matching every boundary point with the seed spot that is mostly connected to it.(4)Ellipse fitting.Regard all boundary points that are matched with the same seed spot as members of one colony object,and then restore the whole outline of the object by ellipse fitting.Thirdly,we propose a method based on Circle Hough Transformation(CHT)to segment colony images with grid background.This method consists of three parts.The first part is to acquire relevant characteristic value of colony targets by CHT.The second part is to obtain two categorizations of candidate targets through the characteristic values obtained above,and the last part is to build a new gray image using the candidate target,followed by the Otsu algorithm to obtain finally result.Fourthly,two proposed algorithms are implemented with Matlab programming language in the Windows 7 environment.Some existing adhesion target segmentation algorithms are also analyzed.Results show that the proposed segmentation algorithm for those colony images with overlapping objects has competitive performance with some existing methods.Moreover,the proposed segmentation algorithm for those colony images with grid background can also deal with homologous colony images.Finally,we develop a colony image segmentation software by integrating some existing colony image segmentation algorithms and the two proposed methods.This software is implemented with C++ programming language in the Windows 7 and Visual Studio 2015 environment,which contains seven functional modules,i.e.,acquisition of colony images,conversion of pattern,delimitation of effective area and setting of parameters,segmentation of colony image,manually correction,display and save of information,zoom in/out of focus areas.