Boundary Preserving Dense Local Regions Jaechul Kim and
Boundary Preserving Dense Local Regions Jaechul Kim and Kristen Grauman Univ. of Texas at Austin Local feature detection A crucial building block for many applications Image retrieval Object recognition Image matching Key issue: How to detect local regions for feature extraction?
Related work Interest point detectors e.g., Matas et al. (BMVC 02), Jurie and Schmid (CVPR 04), Mikolajczyk and Schmid (IJCV 04) Dense sampling e.g., Nowak et al. (ECCV 06) Segmented regions and Superpixels e.g., Ren and Malik (ICCV 03) , Gu et al. (CVPR 09), Todorovic and Ahuja (CVPR 08), Malisiewicz and Efros (BMVC 07), Levinshtein et al. (ICCV 09) Hybrid e.g., Tuytelaars (CVPR 10), Koniusz and Mikolajczyk (BMVC 09) What makes a good local feature detector?
Desired properties: - Repeatable - Boundary-preserving - Distinctively shaped Existing methods lack one or more of these criteria, e.g., Lack repeatability Segments Lack distinctive shape, straddle boundaries Dense sampling Interest points Our idea:
Boundary Preserving Local Regions (BPLRs) Boundary preserving, dense extraction Segmentation-driven feature sampling and linking Repeatable local features capturing objects local shapes Approach: Overview Sampling elements Initial elements for each segment are sampled based on distance transform of the segment A segment Sampled elements Linking elements A single graph structure reflecting main shapes and segment layout Min. spanning tree
Grouping elements Grouping neighboring elements into BPLR Neighbor elements BPLR Sampling Linking Grouping Approach: Sampling x x
An element Zoom-in view Input image Segment Sampled elements Distance Dense regular transform grid from all segments
Sampling Linking Grouping Approach: Linking Minimum spanning tree Sampled elements locations (i.e., elements centers) Global linkage structure
Sampling Linking Grouping Role of spanning tree linkage Min spanning tree prefers to link closer elements + Due to distance transform-based sampling same-segment elements more likely linked Multiple sampling
Due to multiple segmentations elements in overlapping segments more likely linked Sampling Linking Grouping Approach: Grouping Descriptor Example BPLRs Referencedetections
elementsoflocation (Subset shown for visibility) Topological Euclidean Zoom-in neighbor neighbor view Neighbor BPLR elements Reference elements location Euclidean Topological neighbor neighbor
elements elements Intersection of topology and location and Euclidean Euclidean neighbor neighbor Example matches of BPLRs Leak object boundary Experiments
20-200 segments ~7000 BPLRs in 400 x 300 image 2-5 seconds to extract BPLRs per an image PHOG + gPb descriptor used Baselines: Tasks: Repeatability Localization Foreground segmentation Object classification Dense sampling (+ SIFT) MSER (+ SIFT)  Semi-local regions (+ SIFT) [2,3] Segmented regions (+ PHOG)  Superpixels 
 Matas et al., BMVC 02.  Quack et al., ICCV 07.  Lee and Grauman, IJCV 09.  Arbelaez et al., CVPR 09.  Ren and Malik, ICCV 03. Example feature extractions Proposed Segmented BPLRs regions (Subset shown for visibility) Superpixels Interest regions
(MSERs) Dense sampling Repeatability for object categories Bounding Box Hit Rate False Positive Rate [Quack et al. 2007] Test image Applelogo Giraffe Bottle Swan
Mug Train images True match False positive Comparison to baseline region detectors on ETHZ shape classes Localization accuracy Bounding Box Overlapping Score Recall Applelogo Compute overlapping score by projecting the training
exemplars bounding box into the test image Giraffe Bottle Swan Mug Comparison to baseline region detectors on ETHZ shape classes Localization accuracy Test image Database images with best matches to test BPLRs
Foreground segmentation Replacing superpixels with BPLRs in GrabCut segmentation Approach BPLR + GrabCut (Ours) Superpixel + GrabCut Superpixel ClassCut (Alexe et al., ECCV 10) Superpixel Spatial Topic Model (Cao et al., ICCV 07) Accuracy(%) 85.6 81.5 83.6 67.0 Foreground segmentation in Caltech-28 dataset
Object classification Nearest-neighbor results on Caltech-101 benchmark Feature Accuracy(%) BPLR + PHOG (Ours) Dense + SIFT Segment + PHOG Dense + PHOG 61.1 55.2 37.6 27.9 Comparison of features using the same Nave
Bayes NN [Boiman et al. 2008] classifier. Conclusion Dense local detector that preserves object boundaries Capture objects local shape in a repeatable manner Feature sampling and linking driven by segmentation Generic bottom-up extraction Code available: http://vision.cs.utexas.edu/projects/bplr/bplr.html
presenta practicas para salir en cuerpo astral concientemente v.m.samael aun weor. sabiduria gnostica mantran faraon faaaa-rrraaaa-oooonnnn chac-mool mantram rusti ruuuuuusssssstiiiiii mantram tairerere tai-re-re-re mantram silvo, e, r, s silvo e r s cerro de chapultepec chapul chapulin o grillo tepec...
Utilitarian Ethics. The utilitarian ethical theory is founded on the ability to predict the consequences of an action. One benefit of this ethical theory is that the utilitarian can compare similar predicted solutions and use a point system to determine...
The maximum cab angle on a hill side. Slope is 55°. Watch the following videos to see what this machine can do. SAFETY FACTS ABOUT THE FELLER-BUNCHERWHAT YOU NEED TO KNOW1-stay back 300 feet from this machine2- this machine throws...
Historical Research ... cannot ensure representation of the sample Action Research The word "academic" is a synonym for irrelevant. ... with this statement? Why or why not? In your opinion, what is the role of academics, or outsiders, in PAR?...
In a parallel circuit, each load has its own path for electricity. Create a parallel circuit using the materials at your table. (the switch is optional) ... Explain the technical issues that could arise and cause an incomplete path.
returns an unsigned integer value that is the beginning position for the first occurrence of a particular substring within the string . The . substring. argument can be a . string . constant, a . string. expression, or a ....
They consult with students, campus stakeholders, community leaders, legislators, and alumni. Bobcats forever, their work continues even after serving their term, at which point they join the Thomas Ewing Society. ... Chair Bryon Carley Brenda Dancil-jones Alissa Galford. Lyndsay Markley...
VTAC is the central body for applications to Victorian universities, TAFEs, and Independent Tertiary Colleges. VTAC calculates the Australian Tertiary Admission Rank (ATAR) for Victorian Year 12 students. Connect with VTAC to receive updates, tips, and timely reminders:
Ready to download the document? Go ahead and hit continue!