Improved Contour and Texture-Based Object Segmentation
The objective of this work is the detection of objectclasses.An improved method is used for objectdetection and segmentation in real-world multiple-object scenes.It has two stages.In the first stage thismethod develops a novel technique to extract class-discriminative boundary fragments and the texturefeatures near the boundary,and then boosting is usedto select discriminative boundary fragments(weakdetectors)to form a strong Boundary-Fragment-Model detector.An Appearance model is built withthose entire detectors and the texture features.In thesecond stage,the boundary fragment and the texturefeatures and used to complete detection.To the end,anew fast cluster algorithm is used to deal with thecentroid image.The generative aspect of the model isused to determine an approximate segmentation.Inaddition,we present an extensive evaluation of ourmethod on a standard dataset and compare itsperformance to existing methods from the literature.As is shown in the experiment,our method outperformspreviously published methods with the overlap part ofthe object in multiple-object scene.
Lin Kezheng Li Xinyuan Liu Pie
Harbin University of Science and Technology,Harbin 150080,China
国际会议
厦门
英文
1146-1151
2008-11-17（万方平台首次上网日期，不代表论文的发表时间）