基于背景分布空间构建的图像显著性物体检测-电路与系统专业论文.docxVIP

优秀毕业论文 精品参考文献资料 基于背景分布空间构建的图像显著性物体检测模型的检测结果更接近人的主观感受。 基于背景分布空间构建的图像显著性物体检测 模型的检测结果更接近人的主观感受。 关键词：显著性检测；背景分布空间；特征分解；马氏距离 万方数据 AbstractAbstract Abstract Abstract Salient object detection ailns to detect and segment objects or regions tllose receiVe掣eat a：ttention of hmlan洫natural scenes．Tllis topic h嬲attraCted a lot of focused research，because it helps computer harness complex Vision pmblems锄d results in marly applications in tlle field of cornputer Vision， such嬲scene uI-derStalld证g，objeCt detec廿on and recogIlition，洫age segmemation and iInage edith培，m趴ipmatiIlg，etC．111 this p印er，we propose趾emcient bo讹m-up觑吼ework for salient objeCt detection upon me imler statistics of iInage patches．The main contributions ofthe proposed algoritllm are嬲follows： FirsUygiVen the baCk掣ouIldIless prior，me patches 10cated in hage border are diVided into four groups based on me neighbo血g colllle以V咄 111en， eigendecomposition is utilized to create four baCk伊oulld-baSed distribution spaces (SBDs)to model me back铲ound，iIl wmch a more appropriate me仃iC_Mahal姐0bis distallce is quoted to delicately me鹊ure t11e saliency of eVe巧洫age patch away丘．om 恤back孕．0und． Secondly； a noVel sInootlling metllod based on t11e geodesic disCarlCe of supe印ixels is proposed．Combing wim t11e bouIlda巧coIlllectiVity desc咖tor’t11e up-s锄叩liIlg memod upon tlle geodesic distallce of supe印i)【els is used to smootll the coarse saliency m印．Twro mahl a(1vamages are 1isted below：(a)The bounda巧of廿1e salient object collld be preserved、Vell．(b)111is smoo龇g me也0d c肌help 11iglllight tlle salient object u11ifomly，because me geodesic distance of supe印ixels c觚expose t11e regional homogenei够iIl a way． 7111irdly，to be more discmiIlatiVe，t11e smootlled saliency m印is f啪1er erlllanced irIto me posterior probabili锣map witllin Bayesian perspectiVe．nle prior probabilit)r map is defmed aS t11e smootlled saliency m印，趴d mc likelmod is detemlined in tlle ex仃acted salient region arld back铲ound region砌ch are coarsely estimated by biIl耐ziIlg tlle s