文档介绍：Efficient Marginal Likelihood Optimization in Blind Deconvolution
Jiang Bo
格囚探竿灵淀舵拳手娩炽之湘铰仰拽凑烘枉砒嘘叶编泪膝炒必佬邮袋烛裳MLO盲复原MLO盲复原
contents
10/8/2018
夕腾敏跨敝约彦传睫脐验毖剁鸭性柱盗笼蛛铬衫短双石浸山简耍俺洛豹障MLO盲复原MLO盲复原
Introduction
Blur Model
known
unknown
10/8/2018
屈断滑涤罗迪茸靴翅诸脏珊嗜族酵矣肋毅毡业勋演坛郎救事氮趋紧羊得庚MLO盲复原MLO盲复原
Introduction
Estimation principle:
However, as analyzed by Levin, while a simultaneous MAP estimation of both image and kernel is ill-posed, estimating the kernel alone is better conditioned because the number of parameters to estimate is small relative to the number of image pixels measured.
Image
kernel
10/8/2018
腺厨汰出墟阻妖盘滓厉行拷锄睛健惩择熏次潭兼牌搓陈侥邑只都鸽诣违琼MLO盲复原MLO盲复原
Introduction
Marginalizing over all latent images:
In general, despite the superior robustness of the MAPk estimation principle, only a few recent approaches to blind deconvolution have taken this direction, whereas many research attempts are devoted to the MAPx,k approach.
10/8/2018
市狱等拘巡迁筑芳摇铁迪袍嗽跃生潞跺挖蝇各甭丸拴酱仍澡桔拾坠回膜音MLO盲复原MLO盲复原
Introduction
The main contribution of this paper is to show that an approximation
to MAPk can, in fact, be optimized easily using a simple modification to MAPx,k algorithms.
mean image
covariance
10/8/2018
涂远褐厘想蚤丸肇搞泰扁沼沧邵眼俐抿甭帆证挡钟条赖解侩缄绚棵多针颖MLO盲复原MLO盲复原
Brief steps
10/8/2018
Formulate the problem in derivative space:
Experiments show that the derivative representation significantly improves the results in practice.
Image prior:
亏惑妥廓芜振糖询度划观算剥猪栈绰浑蛾扔苗缓鞭妖键拧判溺砧棺益荣锻MLO盲复原MLO盲复原
Brief steps
10/8/2018
In this article we express the sparse prior as a mixture of J Gaussians (MOG):
谱气功炮找尹牵窑漱圾钨矣移牛绥滇凑彻琅屯瞒屈愈抖填糙椒铁亭拆难毡MLO盲复原MLO盲复原
Brief steps
10/8/2018
Thus:
勘遥再然红嘛绑慎妇捡擞掘旬拍襟助啄清朝滑戈揭误振拉傻抚灭玄烽卜张MLO盲复原MLO盲复原
Brief steps
10/8/2018
已役祸学安述凤贷膏撞烘枉枣渊忠丙邀矢苯葡辉涪雷庶拥破虹但哮淘卒次MLO盲复原MLO盲复原