1 / 10

Random Walks on Graphs to Model Saliency in Images

Random Walks on Graphs to Model Saliency in Images. Viswanath Gopalakrishnan Yiqun Hu Deepu Rajan School of Computer Engineering Nanyang Technological University, Singapore 639798 CVPR2009 Reporter: Chia-Hao Hsieh Date: 2010/05/04. Outline. Introduction Methods Experimental results.

iokina
Download Presentation

Random Walks on Graphs to Model Saliency in Images

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Random Walks on Graphs to Model Saliency in Images ViswanathGopalakrishnanYiqunHuDeepuRajan School of Computer Engineering Nanyang Technological University, Singapore 639798 CVPR2009 Reporter: Chia-Hao Hsieh Date: 2010/05/04

  2. Outline • Introduction • Methods • Experimental results

  3. Introduction • Salient region detection based on Markov random walks • Combine global/local property • Seeded salient region identification

  4. Methods • Graph representation • Nodes, Edges • Finding the most salient node

  5. Graph Representation • Node: patches of size 8x8 on the image • Edge: connection between the nodes • Feature: color/orientation • Cb, Cr • Orientation histogram entropy at 5 different scales HP(θi) is the histogram value of the ith orientation bin corresponding to the orientation θi Node feature vector

  6. Finding the Most Salient Node • Most salient: globally pop-out (isolated) • A random walker take more time to reach • If on a compact object, a random walker take less time to reach on a k-regular graph

  7. Background Nodes • Inhomogeneous background • Goal: Capture as much of these variations as possible by locating at least one background node in each of such regions • Maximizing the distance to the most salient node, also maximizing the distance between background nodes

  8. Seeded Salient Region Extraction • If global only • Erroneously classify to salient region • If local only • Spatially close to a salient node • Linearly combine global and local feature

  9. Experimental results

  10. Experimental results Comparison of precision, recall, f-Measure Failure examples

More Related