Interested Object Segmentation Method For Automatically Extracting Marker Points

Digital image processing is a traditional and efficient technique by means of computer to analyze,process,recognize and understand the image,also known as computer image processing.It has been ideally applied in so many momentous areas including machine vision,medical and aerospace.Image segmentation is a crucial technique of digital image processing,and the advantages and disadvantages of the results and the significance of the segmentation of the region directly affect the follow-up work..An effective image segmentation algorithm proposed and the realization of the development of all walks of life has important practical significance.In recent years,It has been arisen a number of technologies which combined the related theory of deep learning and computer vision,Which convolution neural network in many areas have been successful.It is normally based on fetching the characteristic from the image in the traditional image segmentation algorithm.Firstly,divide the image into different regions by use of relevant methods,and then post-process operations(classify the regions,merge the parts,etc.)to acquire meaningful segmentation results ultimately.It is a cumbersome and complicated process that has a lot of improvement space.Convolution neural network is able to extract the image features automatically,so that can be suitable for image processing.Therefore,it has become an important research direction for Improvement of Image Segmentation by CNN Method.The fully convolution neural network(FCN)has been proposed as a novel method in the field of image semantic segmentation.The semantic segmentation operation of FCN algorithm has obvious advantages compared with the traditional segmentation method,such as it is property for the image segmentation through the alterations of the fully connection layers.However,the shortcoming we must to be solved is accuracy because it is lack the characterization of the details of the interested object only separate the approximate region of the interested object.It has a pivotal position that a more detailed segmentation of the accurate target area in the image for the process of image understanding and analysis.It is the urgent issue to tackle that the Segmentation results are not accurate enough in FCN.This article has fully researched the related work,and discussed the application of fully convolution neural network in image segmentation algorithm.The modified method combined with the traditional watershed algorithm further operates the segmentation results from FCN to solve the problem that segmentation of the fully convolution neural network is not enough and the edge detail of the target object is neglected.using the watershed algorithm which has good characteristics of weak edge.And using the approximate region of interest is segmented by the fully convolution neural network(FCN)as a priori knowledge.improve the acquisition method of marker points to refine the target edge and also get better segmentation results.Finally,the superiority of the method we proposed is reflected by comparing with the results of FCN and other segmentation algorithms.