Radial Basis Functions

Similar documents
Machine Learning 9. week

S1 Note. Basis functions.

Computer Animation and Visualisation. Lecture 4. Rigging / Skinning

Point Cloud Surface Representations

Feature Reduction and Selection

GSLM Operations Research II Fall 13/14

Barycentric Coordinates. From: Mean Value Coordinates for Closed Triangular Meshes by Ju et al.

Classification / Regression Support Vector Machines

y and the total sum of

Face Recognition University at Buffalo CSE666 Lecture Slides Resources:

Programming in Fortran 90 : 2017/2018

LECTURE : MANIFOLD LEARNING

Harmonic Coordinates for Character Articulation PIXAR

NAG Fortran Library Chapter Introduction. G10 Smoothing in Statistics

Very simple computational domains can be discretized using boundary-fitted structured meshes (also called grids)

Calibrating a single camera. Odilon Redon, Cyclops, 1914

Hermite Splines in Lie Groups as Products of Geodesics

An Accurate Evaluation of Integrals in Convex and Non convex Polygonal Domain by Twelve Node Quadrilateral Finite Element Method

Subspace clustering. Clustering. Fundamental to all clustering techniques is the choice of distance measure between data points;

Range images. Range image registration. Examples of sampling patterns. Range images and range surfaces

Lecture #15 Lecture Notes

CS 534: Computer Vision Model Fitting

2x x l. Module 3: Element Properties Lecture 4: Lagrange and Serendipity Elements

Analysis of 3D Cracks in an Arbitrary Geometry with Weld Residual Stress

Lecture 5: Multilayer Perceptrons

Interpolation of the Irregular Curve Network of Ship Hull Form Using Subdivision Surfaces

Angle-Independent 3D Reconstruction. Ji Zhang Mireille Boutin Daniel Aliaga

AC : TEACHING SPREADSHEET-BASED NUMERICAL ANAL- YSIS WITH VISUAL BASIC FOR APPLICATIONS AND VIRTUAL IN- STRUMENTS


A Robust Method for Estimating the Fundamental Matrix

Support Vector Machines

SIGGRAPH Interactive Image Cutout. Interactive Graph Cut. Interactive Graph Cut. Interactive Graph Cut. Hard Constraints. Lazy Snapping.

Smoothing Spline ANOVA for variable screening

12/2/2009. Announcements. Parametric / Non-parametric. Case-Based Reasoning. Nearest-Neighbor on Images. Nearest-Neighbor Classification

Wavefront Reconstructor

5 The Primal-Dual Method

LECTURE NOTES Duality Theory, Sensitivity Analysis, and Parametric Programming

Solutions to Programming Assignment Five Interpolation and Numerical Differentiation

SLAM Summer School 2006 Practical 2: SLAM using Monocular Vision

Fitting & Matching. Lecture 4 Prof. Bregler. Slides from: S. Lazebnik, S. Seitz, M. Pollefeys, A. Effros.

Recognizing Faces. Outline

Support Vector Machines

Multi-Resolution Geometric Fusion

The Theory and Application of an Adaptive Moving Least. Squares for Non-uniform Samples

Outline. Discriminative classifiers for image recognition. Where in the World? A nearest neighbor recognition example 4/14/2011. CS 376 Lecture 22 1

Model-Based Bundle Adjustment to Face Modeling

Smooth Approximation to Surface Meshes of Arbitrary Topology with Locally Blended Radial Basis Functions

Exercises (Part 4) Introduction to R UCLA/CCPR. John Fox, February 2005

Machine Learning. Support Vector Machines. (contains material adapted from talks by Constantin F. Aliferis & Ioannis Tsamardinos, and Martin Law)

Reading. 14. Subdivision curves. Recommended:

NUMERICAL SOLVING OPTIMAL CONTROL PROBLEMS BY THE METHOD OF VARIATIONS

Simplification of 3D Meshes

Outline. Seamless Image Stitching in the Gradient Domain. Related Approaches. Image Stitching. Introduction Related Work

Quality Improvement Algorithm for Tetrahedral Mesh Based on Optimal Delaunay Triangulation

Vanishing Hull. Jinhui Hu, Suya You, Ulrich Neumann University of Southern California {jinhuihu,suyay,

Discriminative Dictionary Learning with Pairwise Constraints

SUMMARY... I TABLE OF CONTENTS...II INTRODUCTION...

Problem Set 3 Solutions

Denoising Manifold and Non-Manifold Point Clouds

LEAST SQUARES. RANSAC. HOUGH TRANSFORM.

Prof. Feng Liu. Spring /24/2017

Using Radial Basis Functions to Solve Geodesics Equations for Body Measurements *

Life Tables (Times) Summary. Sample StatFolio: lifetable times.sgp

Robust Surface Reconstruction from Orthogonal Slices

INF 4300 Support Vector Machine Classifiers (SVM) Anne Solberg

Outline. Self-Organizing Maps (SOM) US Hebbian Learning, Cntd. The learning rule is Hebbian like:

Conditional Speculative Decimal Addition*

Inverse Kinematics (part 2) CSE169: Computer Animation Instructor: Steve Rotenberg UCSD, Spring 2016

On the diameter of random planar graphs

Surface reconstruction from sparse data by a multiscale volumetric approach

Kent State University CS 4/ Design and Analysis of Algorithms. Dept. of Math & Computer Science LECT-16. Dynamic Programming

Computer Graphics. - Spline and Subdivision Surfaces - Hendrik Lensch. Computer Graphics WS07/08 Spline & Subdivision Surfaces

Research on laser tracker measurement accuracy and data processing. Liang Jing IHEP,CHINA

CS 231A Computer Vision Midterm

Multi-stable Perception. Necker Cube

Dijkstra s Single Source Algorithm. All-Pairs Shortest Paths. Dynamic Programming Solution. Performance. Decision Sequence.

Solitary and Traveling Wave Solutions to a Model. of Long Range Diffusion Involving Flux with. Stability Analysis

An efficient iterative source routing algorithm

Page 0 of 0 SPATIAL INTERPOLATION METHODS

Structure from Motion

Distance based similarity measures of fuzzy sets

Outline. Midterm Review. Declaring Variables. Main Variable Data Types. Symbolic Constants. Arithmetic Operators. Midterm Review March 24, 2014

Active Contours/Snakes

Mathematics 256 a course in differential equations for engineering students

Machine Learning. K-means Algorithm

An Optimal Algorithm for Prufer Codes *

Generalized, Basis-Independent Kinematic Surface Fitting

A Newton-Type Method for Constrained Least-Squares Data-Fitting with Easy-to-Control Rational Curves

A Hybrid System for Three-Dimensional Objects Reconstruction from Point-Clouds Based on Ball Pivoting Algorithm and Radial Basis Functions

Dijkstra s Single Source Algorithm. All-Pairs Shortest Paths. Dynamic Programming Solution. Performance

Problem Definitions and Evaluation Criteria for Computational Expensive Optimization

2-Dimensional Image Representation. Using Beta-Spline

Learning an Image Manifold for Retrieval

Course Introduction. Algorithm 8/31/2017. COSC 320 Advanced Data Structures and Algorithms. COSC 320 Advanced Data Structures and Algorithms

Fair Triangle Mesh Generation with Discrete Elastica

Understanding K-Means Non-hierarchical Clustering

Explicit Formulas and Efficient Algorithm for Moment Computation of Coupled RC Trees with Lumped and Distributed Elements

BFF1303: ELECTRICAL / ELECTRONICS ENGINEERING. Direct Current Circuits : Methods of Analysis

Assignment # 2. Farrukh Jabeen Algorithms 510 Assignment #2 Due Date: June 15, 2009.

Image Alignment CSC 767

Transcription:

Radal Bass Functons Mesh Reconstructon Input: pont cloud Output: water-tght manfold mesh Explct Connectvty estmaton Implct Sgned dstance functon estmaton Image from: Reconstructon and Representaton of 3D Objects wth Radal Bass Functons by Carr et al. Image from: Reconstructon and Representaton of 3D Objects wth Radal Bass Functons by Carr et al. Sgned Dstance Functon Surface Reconstructon F(x) = 0 Surface F(x) < 0 nsde F(x) > 0 outsde Image from: www.farfeldtechnology.com Gven a set of ponts P= { p, p,..., p } p R 1 2 Fnd a manfold surface S whch approxmates P n 3

Implct Surface Reconstructon Gven a set of ponts P= { p, p,..., p } p R 1 2 Fnd a manfold surface S whch approxmates P n 3 Implct Surface Reconstructon 1) Fnd sgned dstance functon d(x) from pont cloud 2) Evaluate d(x) on unform 3D grd S s zero level-set of a sgned dstance functon d S = { x d( x) = 0} 3) Extract mesh wth Marchng Cubes Fndng the Dstance Functon Many optons Dstance from tangent planes Radal bass functons Posson equaton... Scattered Data Interpolaton Problem Gven N ponts Reconstruct a functon S(x), such that Constrants S(x) contnuous S(x) s smooth ( x, f ) Sx ( ) = f

Radal Bass Functons Soluton Assume S(X) s weghted sum of bass functons ( ) Sx ( ) = wφ x x Radal Bass Functons Zoo Polyharmonc splne k φ ( r) = r log( r), k = 2,4,6,... k ( ), 1,3,5,... φ r = r k = = (,...) Radal Bass Functons Zoo Multquadratc () r φ = r +β 2 2 Gaussan 2 φ () r = e βr Compact support Fndng the Weghts Fnd weghts by solvng Solve a lnear system ( ) f = S( x ) = wφ x x j j j φ w = f, φ =φ x x ( ) j j j

Fndng the Weghts System s sparse when RBFs have compact support Gaussans vs. Polyharmoncs 2 Gaussans φ () r = e βr have compact support But parameter β depends on scale of data ponts requres tunng Non sparse system problematc for large number of ponts Fllng Holes - Extrapolaton Addng a polynomal to the RBF sum Allows to recreate polynomals exactly Improves extrapolaton Addng a Polynomal ( ) S( x) = wφ x x + P( x) P(x) low degree polynomal Wthout polynomal Wth polynomal

Reconstructon usng RBF Lookng for dstance functon, remember? For each pont p n nput we have d( p ) = 0 Use some RBF, solve and get d? Reconstructon usng RBF Problem, RBF system s Mw = 0 Wll get the trval soluton Add off-surface ponts Off-surface Ponts For each pont n data add 2 off-surface ponts on both sdes of surface Use normal data to fnd off-surface ponts Off-surface Ponts P= { p, p,..., p } p R 1 2 d( p ) = 0 d( p + cnˆ ) = c d( p cnˆ ) = c n 3 Reconstructon usng RBF needs orented ponts Image from: Reconstructon and Representaton of 3D Objects wth Radal Bass Functons by Carr et al. Images from: Reconstructon and Representaton of 3D Objects wth Radal Bass Functons by Carr et al.

Off-surface Ponts Estmated normal mght be ambguous Don t add off-surface ponts there Project to whch dstance? Closest surface pont to projected pont should be generatng pont Reducng Number of Centers Not necessarly need all ponts to be centers to reconstruct surface Constant projecton dstances Varyng projecton dstances Images from: Reconstructon and Representaton of 3D Objects wth Radal Bass Functons by Carr et al. Images from: Reconstructon and Representaton of 3D Objects wth Radal Bass Functons by Carr et al. Reducng Number of Centers Greedy approach Start wth a subset of the ponts Repeat Ft RBFs Evaluate error at all ponts If larger than tolerance, and more centers Computatonal Complexty RBFs mght gve dense matrx O(N 2 ) - Storage O(N 3 ) Solvng the matrx O(N) Evaluatng a pont Fast methods exst O(N) Storage O(NlogN) Fttng O(1) - Evaluaton Images from: Reconstructon and Representaton of 3D Objects wth Radal Bass Functons by Carr et al. 544,000 ponts, 80,000 centers err = 0.0005

Examples Images from: Reconstructon and Representaton of 3D Objects wth Radal Bass Functons by Carr et al.