基于强化学习的图像分割算法研究.pdfVIP

计算机科学2007V01．34No．8A 基于强化学习的图像分割算法研究*) to BasedonReinforcement ApproachImageSegmentation Learning 朱亮1 赵波1 高 阳1 汪栋2 张樱2 叶玉坤2 (南京大学计算机软件新技术国家重点实验室 南京210093)1 (中国人民解放军南京军区第八一医院 南京210002)2 flew im- AbstractInthis a methodthatusesreinforeementtO medical paper，weproposeadaptive learningsegment newmethodtO cell inordertOlet learntofinda this segmentcancer ages．Weapply lung images agent optimum thresholdThemethodof basedonreinforcementdefinesthatonestateisthe CO imagesegmentation learning edge ratioof and detection otherstateiSthearea ratioof incidencesegmentedimageedge iesult．出e coincidencesegrnen- andtheresuhof0S兀7 method：AnditdefinesthatactioniSthreshold ted曲．flge imagesegmentation change；A1SO it coincidenceratioof and USesthe defines segmented ir工1．age optimumsegmentation．Reinforcementlearningagent trial-and-errormethodtO1eamthe action reinforcement canusethis optLmumpolicy．The Iearningagent knowledge similar aswellFrom inthelastofthis canseethatthecorrect to other the segment hnages experiments paper，we ratioofthisnewmethodiSabout91％．andthe iS performancegood． Reinforcement Keywords learning，CeUsegmentation，Imageprocessing 习应用到对肺癌细胞图像的分割中。强化学习A— 1 引言 gent(一种自适应的，基于状态一动作一奖赏的学习算 目标识别是图像处理和模式识别领域的研究热 法)自主的选择动作策略，在与环境的交互中获得奖 点，在生物医学领域有着十分广泛的应用