1 / 1

Problem

Heterogeneous Conditional Random Field (HCRF): Realizing Joint Detection and Segmentation of Cell Regions in Microscopic Images Jiyan Pan 1 , Takeo Kanade 1 , and Mei Chen 2

wilbur
Download Presentation

Problem

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. Heterogeneous Conditional Random Field (HCRF): Realizing Joint Detection and Segmentation of Cell Regions in Microscopic Images Jiyan Pan1, Takeo Kanade1, and Mei Chen2 1Carnegie Mellon University, 2Intel Labs Pittsburgh 1 {jiyanpan, tk}@cs.cmu.edu, 2mei.chen@intel.com Results Conclusion • Cell Type • bovine aortic endothelial cells • C2C12 muscle stem cells • For each cell type • 10 images for training • 10 images for testing • Compare HCRF with • separate detection and segmentation • conventional CRF Detect and segment out individual cells in a dense population in microscopic images • The state set is {0,1,…,N}, N is the total number of interest points • The resulting model isunidentifiable • Maximum posterior probability shared by several states • CRF cannot select the correct state assign- ment combination SIFT? No stable spatial structures Sliding window? Cell shapes highly irregular N-cuts? Not discriminative Needs total # of cells Bovine C2C12 Input • Proposed approach: • Extract interest points and features • Classify points into cell or background (detection) • Group points within the same cell (segmentation) • Extend points to regions Conventional CRF Problem Approach Separate Detection/Segmentation Heterogeneous CRF (HCRF) Joint Detection/Segmentation by CRF Separate HCRF Before MAP inference • Give nodes an arbitrary ordering • Restricted propagation rule • Each node propagates its node index in turn • A node neither accepts nor passes on any state greater than its node index CRF • Two critical parameters to tune • Cannot recover from detection errors • No mutual enhancement between detection and segmentation Remaining Unidentifiability Bovine C2C12 After MAP inference Non-maxima suppression rule If a node’s maximum posterior probability is shared by several states, it takes the largest state • Joint detection and segmentation outperforms sep- arate detection and segmentation • Conventional CRF cannot achieve joint detection and segmentation due to unidentifiability • HCRF resolves unidentifiability by heterogeneous st- ate sets and non-maxima suppression rule • HCRF is provably complete, irreducible, unique, and sound

More Related