Matthew Brown University of British Columbia (prev.) Microsoft Research [ Collaborators: † Simon Winder, *Gang Hua, † Rick Szeliski † =MS Research, *=MS.

Slides:



Advertisements
Similar presentations
Distinctive Image Features from Scale-Invariant Keypoints David Lowe.
Advertisements

Summary of Friday A homography transforms one 3d plane to another 3d plane, under perspective projections. Those planes can be camera imaging planes or.
The SIFT (Scale Invariant Feature Transform) Detector and Descriptor
Computer vision: models, learning and inference Chapter 13 Image preprocessing and feature extraction.
TP14 - Local features: detection and description Computer Vision, FCUP, 2014 Miguel Coimbra Slides by Prof. Kristen Grauman.
MIT CSAIL Vision interfaces Approximate Correspondences in High Dimensions Kristen Grauman* Trevor Darrell MIT CSAIL (*) UT Austin…
Gustavo Carneiro The Automatic Design of Feature Spaces for Local Image Descriptors using an Ensemble of Non-linear Feature Extractors.
Computer Vision for Human-Computer InteractionResearch Group, Universität Karlsruhe (TH) cv:hci Dr. Edgar Seemann 1 Computer Vision: Histograms of Oriented.
Image and video descriptors
Instructor: Mircea Nicolescu Lecture 15 CS 485 / 685 Computer Vision.
Discrete-Continuous Optimization for Large-scale Structure from Motion David Crandall, Andrew Owens, Noah Snavely, Dan Huttenlocher Presented by: Rahul.
Generic Object Detection using Feature Maps Oscar Danielsson Stefan Carlsson
Recognising Panoramas
Features-based Object Recognition Pierre Moreels California Institute of Technology Thesis defense, Sept. 24, 2007.
CVR05 University of California Berkeley 1 Familiar Configuration Enables Figure/Ground Assignment in Natural Scenes Xiaofeng Ren, Charless Fowlkes, Jitendra.
Automatic Panoramic Image Stitching using Local Features Matthew Brown and David Lowe, University of British Columbia.
Automatic Image Alignment (feature-based) : Computational Photography Alexei Efros, CMU, Fall 2005 with a lot of slides stolen from Steve Seitz and.
Lecture 1: Images and image filtering
Lecture 11: Structure from motion CS6670: Computer Vision Noah Snavely.
Automatic Image Stitching using Invariant Features Matthew Brown and David Lowe, University of British Columbia.
Scale Invariant Feature Transform (SIFT)
Automatic Matching of Multi-View Images
Evaluation of features detectors and descriptors based on 3D objects P. Moreels - P. Perona California Institute of Technology.
Image Features: Descriptors and matching
1 Invariant Local Feature for Object Recognition Presented by Wyman 2/05/2006.
Automatic Image Alignment (feature-based) : Computational Photography Alexei Efros, CMU, Fall 2006 with a lot of slides stolen from Steve Seitz and.
Lecture 6: Feature matching and alignment CS4670: Computer Vision Noah Snavely.
Accurate, Dense and Robust Multi-View Stereopsis Yasutaka Furukawa and Jean Ponce Presented by Rahul Garg and Ryan Kaminsky.
Summary of Previous Lecture A homography transforms one 3d plane to another 3d plane, under perspective projections. Those planes can be camera imaging.
Overview Introduction to local features
Distinctive Image Features from Scale-Invariant Keypoints By David G. Lowe, University of British Columbia Presented by: Tim Havinga, Joël van Neerbos.
Internet-scale Imagery for Graphics and Vision James Hays cs195g Computational Photography Brown University, Spring 2010.
A Local Adaptive Approach for Dense Stereo Matching in Architectural Scene Reconstruction C. Stentoumis 1, L. Grammatikopoulos 2, I. Kalisperakis 2, E.
Overview Harris interest points Comparing interest points (SSD, ZNCC, SIFT) Scale & affine invariant interest points Evaluation and comparison of different.
Building local part models for category-level recognition C. Schmid, INRIA Grenoble Joint work with G. Dorko, S. Lazebnik, J. Ponce.
Feature Detection and Descriptors
Multi-Output Learning for Camera Relocalization Abner Guzmán-Rivera UIUC Pushmeet Kohli Ben Glocker Jamie Shotton Toby Sharp Andrew Fitzgibbon Shahram.
Classifying Images with Visual/Textual Cues By Steven Kappes and Yan Cao.
Pedestrian Detection and Localization
Why is computer vision difficult?
Lecture 7: Features Part 2 CS4670/5670: Computer Vision Noah Snavely.
Peter Henry1, Michael Krainin1, Evan Herbst1,
Deconstruction: Discriminative learning of local image descriptors Samantha Horvath Learning Based Methods in Vision 2/14/2012.
Overview Introduction to local features Harris interest points + SSD, ZNCC, SIFT Scale & affine invariant interest point detectors Evaluation and comparison.
CSE 185 Introduction to Computer Vision Feature Matching.
Project 3 questions? Interest Points and Instance Recognition Computer Vision CS 143, Brown James Hays 10/21/11 Many slides from Kristen Grauman and.
Local features: detection and description
Scale Invariant Feature Transform (SIFT)
Paper presentation topics 2. More on feature detection and descriptors 3. Shape and Matching 4. Indexing and Retrieval 5. More on 3D reconstruction 1.
Lecture 9 Feature Extraction and Motion Estimation Slides by: Michael Black Clark F. Olson Jean Ponce.
Presented by David Lee 3/20/2006
CSCI 631 – Foundations of Computer Vision March 15, 2016 Ashwini Imran Image Stitching.
Invariant Local Features Image content is transformed into local feature coordinates that are invariant to translation, rotation, scale, and other imaging.
SIFT Scale-Invariant Feature Transform David Lowe
Presented by David Lee 3/20/2006
Lecture 07 13/12/2011 Shai Avidan הבהרה: החומר המחייב הוא החומר הנלמד בכיתה ולא זה המופיע / לא מופיע במצגת.
CS 4501: Introduction to Computer Vision Sparse Feature Detectors: Harris Corner, Difference of Gaussian Connelly Barnes Slides from Jason Lawrence, Fei.
TP12 - Local features: detection and description
Digital Visual Effects, Spring 2006 Yung-Yu Chuang 2006/3/22
Local features: detection and description May 11th, 2017
Paper Presentation: Shape and Matching
Digit Recognition using SVMS
Modeling the world with photos
Cheng-Ming Huang, Wen-Hung Liao Department of Computer Science
Features Readings All is Vanity, by C. Allan Gilbert,
CS598:Visual information Retrieval
Rob Fergus Computer Vision
Interest Points & Descriptors 3 - SIFT
Image Stitching Computer Vision CS 678
KFC: Keypoints, Features and Correspondences
Presentation transcript:

Matthew Brown University of British Columbia (prev.) Microsoft Research [ Collaborators: † Simon Winder, *Gang Hua, † Rick Szeliski † =MS Research, *=MS Live Labs]

 Panoramic Stitching Digital Image Pro, Windows Live Photogallery, Expression, HDView  3D Modelling Photosynth  Virtual Earth Location Recognition  Image Search Lincoln [ yellow = product, white = technology preview, grey = research ]

[ ]

[ Slide credit: Noah Snavely] Scene reconstruction Photo Explorer Input photographs Relative camera positions and orientations Point cloud Sparse correspondence [ ]  Photosynth is based on Photo Tourism [Snavely, Seitz, Szeliski SIGGRAPH 2006 ]  Photo Tourism uses SIFT for correspondence

[ Seitz et al CVPR 2006, Goesele et al ICCV 2007 ]

[ Photo Tourism – Snavely, Seitz, Szeliski - SIGGRAPH 2006 ] 3D Point Cloud

[ Photo Tourism – Snavely, Seitz, Szeliski - SIGGRAPH 2006 ] 3D Point Cloud

3D Point Cloud [ Photo Tourism – Snavely, Seitz, Szeliski - SIGGRAPH 2006 ]

3D Point Cloud [ Photo Tourism – Snavely, Seitz, Szeliski - SIGGRAPH 2006 ]

 † = for simplicity + efficiency  * = measured by ROC curve  Q: Form of the descriptor function f(.)? Find a function of a local image patch descriptor = f ( ) s.t. a nearest neighbour classifier † is optimal*

Algorithm Normalized Image Patch Descriptor Vector Gradients Quantized to k Orientations Normalize Summation [ SIFT – Lowe ICCV 1999 ]

Algorithm Normalized Image Patch Descriptor Vector Gradients Quantized to k Orientations Normalize (plus PCA) Summation [ GLOH – Mikolajzcyk Schmid PAMI 2005 ]

Algorithm Normalized Image Patch Descriptor Vector Create Edge Map Normalize Summation [ Shape Context – Belongie Malik Puzicha NIPS 2000 ]

Algorithm Normalized Image Patch Descriptor Vector Feature Detector Normalize Summation TSN [ Geometric Blur – Berg Malik CVPR 2001 ]

Normalized Image Patch Descriptor Vector TSN Parameters  Propose a framework for descriptor algorithms  Learn parameters to find best performance  Train on a ground truth data set based on accurate 3D matches

Normalized Image Patch (w x h) Descriptor Vector TSN  Transformation block Local gradients Steerable filters Isotropic filters  Haar wavelets  Local classifier  Quantized intensities (w x h x k)  Output: one length k vector per source pixel

Normalized Image Patch (w x h) Descriptor Vector SNT (w x h x k) (m x k)  Spatial summation block with m regions  Output: m length k vectors S1 S2 S3 S4

Normalized Image Patch (w x h) Descriptor Vector SNT (w x h x k) (m x k)  Normalization Block Unit normalization SIFT normalization with clipping

STN

S2T1aN2 Parameters Training Pairs Incorrect Match % Correct Match % Update Parameters (Powell) Descriptor Distances

S2T1aN2 Parameters Test Pairs Incorrect Match % Correct Match % Final Error Rate Descriptor Distances 95%

 Polar lattice S2 always has lower error rate than rectangular S1  Gradient and DOG with S2 beat our SIFT reference (4% vs 6% error)

 Steerable filters produce great results if phase information is kept

 SIFT normalization is important  Best result: 4 th order steerable filters with phase information combined with polar S4-25 Gaussian summation block (2% error vs SIFT at 6%)  Very large numbers of dimensions

w PCA

w LDA

 LDA on pixels ≈ SIFT (6%)  PCA gave small improvement  Normalised patches Gradient patches

Effect of # of Training Pairs  LDA on pixels ≈ SIFT (6%)  PCA gave small improvement  Need ~100,000 training examples

 LDA on T1-T3 < 4.5%  Optimal #dimensions ~20-30  Post-normalisation important T1 T3

LDA using T blocks T1–T4  LDA on T1-T3 < 4.5%  Optimal #dimensions ~20-30  Post-normalisation important

LDA using CVPR 07 descriptors  Overall best results  #dimensions reduced from 100’s to 10’s  Need more challenging dataset!

Algorithm Normalized Image Patch Descriptor Vector Feature Detector Normalize Summation TSN “complex” “simple”

 Used learning to obtain good descriptors  Achieved error rates 1/3 of SIFT  Produced useful ground truth data set  Future Work  Use multi-view stereo ground truth  Multi-level simple-complex architecture  + non-parametric T blocks  Learn interest point detectors [ refs: 1) Winder, Brown CVPR ) Hua, Brown, Winder ICCV 2007 ]

[ ]