Research On The Adhesive Circle-like Image Segmentation And Statistic Algorithms Based On Computer Vision

Circle-like objects exist in every corner of our lives, such as medical cells,animal eggs, fruits, downward head image, logs’ cross sections, grain particles and soon. It has become one of the most important research topics in the field of computervision that using digital image processing technology to extract the circle-like targetsfrom the background, and dividing them into some single areas with integrity edges,as well as counting the numbers of the targets in the image and the feature parametersof each target (e.g. roundness, area, circumference, secondary moment, moment,entropy, color etc.) is, which has an important significance in practical application.Generally, there exist various degrees of noise or noise regions in the imagesobtained from nature, meanwhile, the circle-like objects in images are irregular inshape and unequal in size, and most of them are clusters with severe adhesions. Allthese factors described above make it more difficult that the circle-like imagesegmentation is utilized in practical application. The crux of circle-like imagesegmentation is how to avoid the interference of noise effectively and at the sametime separate these adhesion targets successfully. In this paper, aiming at theproblems of adhesive circle-like image segmentation, the algorithms of imagepreprocessing and segmentation based on distance transformation based watershedalgorithm have been studied and improved as well.For the circle-like image segmentation with regular adhesion in simplebackground, a watershed algorithm based on distance map reconstruction is presentedin this paper, in which the top/bottom hat transformation is innovatively introduced.The top/bottom hat transformation can enhance the contrast between background andtargets effectively, which make the separation of adhesive targets simple. In thewatershed segmentation algorithm, the local maximum point is firstly extracted fromdistance image and optimized; and then reconstruct distance image based on theoptimized maximum point; in the last step, watershed the reconstructed distance mapto get the segmented image. The proposed algorithm can separate the adhesive targetswell and suppress the over-segmentation in watershed algorithm effectively; moreover, the algorithm has the universal adaptability.For the adhesive circle-like image segmentation in complex background, thispaper presents a segmentation algorithm based on adaptive region merging andwatershed, in which the region merging algorithm can select different similaritycriteria automatically according to the size of the adjacent areas. The algorithm canextract the circle-like targets from complex background and separate the adhesivetargets with irregular shape. Besides, simulation results show that the algorithm hasgood adaptabilityThe algorithm of the mark counting and characteristic parameters calculation isstudied. A circle-like image segmentation and analysis system is constructed based onthe segmentation and statistical algorithm proposed in this paper.