Learning Spatial Prior with Automatically Labeled Landmarks
We propose a method of automatically labelinglandmarks on target images,which are used fortraining a constellation model to recognize generalobject class.First,we randomly sample local features(parts)and generate hierarchical representations ofimages in a similar way to the standard model ofvisual cortex.Second,we pick out a unique location ofeach part among those local maxima in S2 layers by amatching procedure.Third,we model the spatialrelations among parts as a sparse GMRF(GaussianMarkov Random Fields)graph,and learn the links bya lasso-based approach.Object localizatTion in newimages proceeds by maximizing the posterior of anobject observed at a particular configuration.Ourmodel is a thoroughly automatic scheme to performfeature binding.Experimental results on theCalTechlOl database demonstrate that the proposedalgorithm locates the components more precisely andoutperforms the standard model in object detection.
Object class recognition lnvariance Part constellation Hierarchical model GMRF Sparse graph
Jianzhai Wu Zongtan Zhou Li Zhou Dewen Hu
Department of Automatic Control,College of Mechatronics and Automation,National University of Defense Technology,Changsha,Hunan,410073,China.
国际会议
厦门
英文
1191-1197
2008-11-17（万方平台首次上网日期，不代表论文的发表时间）