concurrent object recognition and segmentation by graph partitioning-slide.pdfVIP

下载本文档

1
0
约3.91千字
约 17页
2017-04-09 发布于江苏
举报
版权申诉

concurrent object recognition and segmentation by graph partitioning-slide.pdf

1、有哪些信誉好的足球投注网站（book118）网站文档一经付费（服务费），不意味着购买了该文档的版权，仅供个人/单位学习、研究之用，不得用于商业用途，未经授权，严禁复制、发行、汇编、翻译或者网络传播等，侵权必究。。
2、本站所有内容均由合作方或网友上传，本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺！文档内容仅供研究参考，付费前请自行鉴别。如您付费，意味着您自己接受本站规则且自行承担风险，本站不退款、不进行额外附加服务；查看《如何避免下载的几个坑》。如果您已付费下载过本站文档，您可以点击这里二次下载。
3、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等，请点击“版权申诉”（推荐），也可以打举报电话：400-050-0827(电话支持时间：9:00-18:30)。
4、该文档为VIP文档，如果想要下载，成为VIP会员后，下载免费。
5、成为VIP后，下载本文档将扣除1次下载权益。下载后，不支持退款、换文档。如有疑问请联系我们。
6、成为VIP后，您将拥有八大权益，权益包括：VIP文档下载权益、阅读免打扰、文档格式转换、高级专利检索、专属身份标志、高级客服、多端互通、版权登记。
7、VIP文档为合作方或网友上传，每下载1次，网站将根据用户上传文档的质量评分、类型等，对文档贡献者给予高额补贴、流量扶持。如果你也想贡献VIP文档。上传文档

concurrent object recognition and segmentation by graph partitioning-slide

Concurrent Object Recognition and Segmentation by Graph Partitioning Stella X. Yu 1, 2 Ralph Gross 1 Jianbo Shi 1 Robotics Institute 1 Carnegie Mellon University Center for the Neural Basis of Cognition 2 YU, GROSS SHI: NIPS’02 VS18 – p.1/17 Why segmentation needs recognition? Why recognition needs segmentation? YU, GROSS SHI: NIPS’02 VS18 – p.2/17 Image segmentation is often object-blind 1. Do not know which regions make up an object. 2. Easily miss object boundaries due to lighting and clutter. YU, GROSS SHI: NIPS’02 VS18 – p.3/17 Object detection is often overwhelmed (Schneiderman, 02): vasc.ri.cmu.edu/demos/faceindex 1. Tradeoff between false positives and detection rate. 2. Constraints in reducing false detection: increase in classifier complexity and training size. YU, GROSS SHI: NIPS’02 VS18 – p.4/17 Characteristics of false positives Arm-1 Leg-1 Head Arm-2 Leg-2 1. Lack of high-level part label compatibility. 2. Lack of low-level image feature saliency. YU, GROSS SHI: NIPS’02 VS18 – p.5/17 Head Arm-2Arm-1 Leg-2Leg-1 B C A Overview of our object segmentation Y U , G R O S S S H I : N I P S ’ 0 2 V S 1 8 – p .6 / 1 7 Representation Graph: G = (V,E,W ) = (nodes, edges, weights) Node set: V = Vpixels ∪ Vpatches Edge set: E = Epixel?pixel ∪ Epatch?patch ∪ Epixel?patch Weights: W = [ A CT C B ] A: pixel-pixel similarity matrix B: patch-patch compatibility matrix C: pixel-patch association matrix YU, GROSS SHI: NIPS’02 VS18 – p.7/17 patches image pixel-patch association object edges segmentation e v a lu a tio n in te g r a tio n YU, GROSS SHI: NIPS’02 VS18 – p.8/17 Computing pixel-pixel similarity A A(1, 3) ≈ 1 A(1, 2) ≈ 0 1 2 3 image filters edge magnitudes A(i, j) = exp ? ?? 1 2σ2e · [ maxt∈(0,1) OE(i ˉ + t · j ˉ ) maxk OE(k ˉ ) ]2?? , k ˉ = location of k. YU, GROSS SHI: NIPS’02 VS18 – p.9/17 Computing patch compatibility and competition B(1, 2) ≈ 1 B(10, 5) ≈ 0 7 and 8 cannot both be parts of the object 1 2 3 4 5 6 7 8 9 10 11 1 2 3 4 5