FFTs in Graphics and Vision. Rotational and Reflective Symmetry Detection

Similar documents
FFTs in Graphics and Vision. Rotational and Reflective Symmetry Detection

CMSC 425: Lecture 4 Geometry and Geometric Programming

Singular Frégier Conics in Non-Euclidean Geometry

1 Properties of Spherical Harmonics

AMS10 HW1 Grading Rubric

= ( 2) = p 5.

Participation Factors. However, it does not give the influence of each state on the mode.

PHYS 301 HOMEWORK #9-- SOLUTIONS

Quaternionic Projective Space (Lecture 34)

Solutions to Problem Set 5

Review of Matrices and Vectors 1/45

Dirac s Hole Theory and the Pauli Principle: Clearing up the Confusion

Central Force Motion Challenge Problems

Scaling Multiple Point Statistics for Non-Stationary Geostatistical Modeling

( ) x y z. 3 Functions 36. SECTION D Composite Functions

Surfaces of Revolution with Constant Mean Curvature in Hyperbolic 3-Space

Categorical Background (Lecture 2)

On Lattice Structure of the Probability Functions on L *

ANALYSIS OF ULTRA LOW CYCLE FATIGUE PROBLEMS WITH THE BARCELONA PLASTIC DAMAGE MODEL

Introduction to Group Theory Note 1

Principles of Computed Tomography (CT)

1 such that v k g. v g. u g

MATH 361: NUMBER THEORY ELEVENTH LECTURE

Elliptic Curves Spring 2015 Problem Set #1 Due: 02/13/2015

A Family of Binary Sequences from Interleaved Construction and their Cryptographic Properties

Computational Acoustic Attenuation Performance of Helicoidal Resonators Comparable to Experiment

Complex Analysis Homework 1

Problem set 6 for Quantum Field Theory course

Hermite subdivision on manifolds via parallel transport

Numerical Linear Algebra

Towards understanding the Lorenz curve using the Uniform distribution. Chris J. Stephens. Newcastle City Council, Newcastle upon Tyne, UK

arxiv: v1 [math.dg] 22 Mar 2017

Notes on Serre fibrations

Lecture 8: The First Fundamental Form Table of contents

Chapter 2 Arithmetic Functions and Dirichlet Series.

Cheng, N. S., and Law, A. W. K. (2003). Exponential formula for computing effective viscosity. Powder Technology. 129(1-3),

LECTURE 6: FIBER BUNDLES

Lecture 1.2 Pose in 2D and 3D. Thomas Opsahl

DIFFERENTIAL GEOMETRY. LECTURES 9-10,

On the relationship between sound intensity and wave impedance

Some Unitary Space Time Codes From Sphere Packing Theory With Optimal Diversity Product of Code Size

Surfaces of Revolution with Constant Mean Curvature H = c in Hyperbolic 3-Space H 3 ( c 2 )

Outline. EECS150 - Digital Design Lecture 26 Error Correction Codes, Linear Feedback Shift Registers (LFSRs) Simple Error Detection Coding

A Geometric Review of Linear Algebra

LINEAR SYSTEMS WITH POLYNOMIAL UNCERTAINTY STRUCTURE: STABILITY MARGINS AND CONTROL

ANALYTIC NUMBER THEORY AND DIRICHLET S THEOREM

CSE 599d - Quantum Computing When Quantum Computers Fall Apart

INJECTIVITY RADIUS AND DIAMETER OF THE MANIFOLDS OF FLAGS IN THE PROJECTIVE PLANES

Various Proofs for the Decrease Monotonicity of the Schatten s Power Norm, Various Families of R n Norms and Some Open Problems

SECTION 12: HOMOTOPY EXTENSION AND LIFTING PROPERTY

8.7 Associated and Non-associated Flow Rules

SECTION 5: FIBRATIONS AND HOMOTOPY FIBERS

Semicontinuous filter limits of nets of lattice groupvalued

Lift forces on a cylindrical particle in plane Poiseuille flow of shear thinning fluids

Statics and dynamics: some elementary concepts

Chater Matrix Norms and Singular Value Decomosition Introduction In this lecture, we introduce the notion of a norm for matrices The singular value de

HEAT AND LAPLACE TYPE EQUATIONS WITH COMPLEX SPATIAL VARIABLES IN WEIGHTED BERGMAN SPACES

Approximating min-max k-clustering

0.1 Practical Guide - Surface Integrals. C (0,0,c) A (0,b,0) A (a,0,0)

Lecture 6. 2 Recurrence/transience, harmonic functions and martingales

Comptes rendus de l Academie bulgare des Sciences, Tome 59, 4, 2006, p POSITIVE DEFINITE RANDOM MATRICES. Evelina Veleva

SYMPLECTIC STRUCTURES: AT THE INTERFACE OF ANALYSIS, GEOMETRY, AND TOPOLOGY

Exercise. Accelerated Math : Monday, May 6, 2013, 7:40 PM Mr. J. Ziarno Ziarno-1st James Ziarno

Dot-Product Steering A New Control Law for Satellites and Spacecrafts

Computer arithmetic. Intensive Computation. Annalisa Massini 2017/2018

TRACES OF SCHUR AND KRONECKER PRODUCTS FOR BLOCK MATRICES

Lecture 8 Optimization

CMSC 425: Lecture 7 Geometric Programming: Sample Solutions

Calculation of gravity due to a vertical cylinder using a spherical harmonic series and numerical integration

Simplifications to Conservation Equations

Chapter 11 Collision Theory

Node-voltage method using virtual current sources technique for special cases

Economics 101. Lecture 7 - Monopoly and Oligopoly

On Nonpermutational Transformation Semigroups with an Application to Syntactic Complexity

Unit 11: Vectors in the Plane

On Closing the Constitutive Gap Between Forming and Crash Simulation

Categories and Natural Transformations

Problem Set. Problems on Unordered Summation. Math 5323, Fall Februray 15, 2001 ANSWERS

Categories and Quantum Informatics: Monoidal categories

Spanning Tree Based State Encoding for Low Power Dissipation

Cybernetic Interpretation of the Riemann Zeta Function

FLOW MEASUREMENT IN CLOSED CONDUIT

Round-off Errors and Computer Arithmetic - (1.2)

ε i (E j )=δj i = 0, if i j, form a basis for V, called the dual basis to (E i ). Therefore, dim V =dim V.

Robustness of classifiers to uniform l p and Gaussian noise Supplementary material

Massachusetts Institute of Technology Department of Electrical Engineering and Computer Science Electric Machinery

PROFIT MAXIMIZATION. π = p y Σ n i=1 w i x i (2)

SOME TRACE INEQUALITIES FOR OPERATORS IN HILBERT SPACES

CSC165H, Mathematical expression and reasoning for computer science week 12

E( x ) [b(n) - a(n, m)x(m) ]

2.9 Dirac Notation Vectors ji that are orthonormal hk ji = k,j span a vector space and express the identity operator I of the space as (1.

44 CHAPTER 5. PERFORMACE OF SMALL SIGAL SETS In digital communications, we usually focus entirely on the code, and do not care what encoding ma is use

Numerical Methods: Structured vs. unstructured grids. General Introduction: Why numerical methods? Numerical methods and their fields of application

ENSC327 Communications Systems 2: Fourier Representations. School of Engineering Science Simon Fraser University

jf 00 (x)j ds (x) = [1 + (f 0 (x)) 2 ] 3=2 (t) = jjr0 (t) r 00 (t)jj jjr 0 (t)jj 3

Topic 7: Using identity types

The Arm Prime Factors Decomposition

GOOD MODELS FOR CUBIC SURFACES. 1. Introduction

MATH 2710: NOTES FOR ANALYSIS

Planar Transformations and Displacements

Transcription:

FFTs in Grahics and Vision Rotational and Relectie Symmetry Detection

Outline Reresentation Theory Symmetry Detection Rotational Symmetry Relectie Symmetry

Reresentation Theory Recall: A rou is a set o elements G with a binary oeration oten denoted such that or all h G the ollowin roerties are satisied: Closure: h G Associatiity: h= h Identity: There exists an identity element G s.t.: = = Inerse: Eery element has an inerse - s.t.: - = - =

Reresentation Theory Obseration : Gien a rou G={ n } or any G the set-theoretic ma that multilies the elements o G on the let by is inertible. The inerse is the ma multilyin the elements o G on the let by -.

Reresentation Theory Obseration : In articular this imlies that the set { n } is just a re-orderin o the set { n }. Or more simly G=G.

Reresentation Theory Obseration : In articular this imlies that the set { n } is just a re-orderin o the set { n }. Or more simly G=G. Similarly the set { - n - } is just a reorderin o the set { n }. Or more simly G - =G.

Reresentation Theory Recall: A Hermitian inner roduct is a ma rom VxV into the comlex numbers that is:. Linear: For all uw V and any real scalar u w u w w w w. Conjuate Symmetric: For all u V w w 3. Positie Deinite: For all V:

Reresentation Theory Obseration : Gien a Hermitian inner-roduct sace V and ien a set o ectors { n } V the ector minimizin the sum o squared distances is the aerae o { n }: n n ar min n V

Reresentation Theory Recall: A unitary reresentation o a rou G on a Hermitian inner-roduct sace V is a ma sends eery element in G to an orthoonal transormation on V satisyin: or all h G. h h that

Reresentation Theory Deinition: We say that a ector V is inariant under the action o G i G sends bac to itsel: or all G.

Reresentation Theory Notation: We denote by V G the set o ectors in V that are inariant under the action o G: V G V G

Reresentation Theory Obseration 3: Note that the set V G is a ector sub-sace o V.

Reresentation Theory Obseration 3: Note that the set V G is a ector sub-sace o V. I w V G then or any G we hae: and w w

Reresentation Theory Obseration 3: Note that the set V G is a ector sub-sace o V. I w V G then or any G we hae: and w And or all scalars and we hae: w w w w

Reresentation Theory Obseration 3: Note that the set V G is a ector sub-sace o V. I w V G then or any G we hae: and w And or all scalars and we hae: So + w V G as well. w w w w

Reresentation Theory Obseration 4: Gien a inite rou G and ien ector V the ector obtained by aerain oer G: Aerae G G G is inariant under the action o G.

Reresentation Theory Obseration 4: To see this let h be any element in G. We would lie to show that h mas the aerae bac to itsel: Aerae G Aerae G h

Reresentation Theory Obseration 4: To see this let h be any element in G. We would lie to show that h mas the aerae bac to itsel: Aerae G Aerae G Exandin the riht hand side we et: h h Aerae G h G G

Reresentation Theory Obseration 4: h Aerae G h G G By the linearity o the reresentation we et: h Aerae G h G G

Reresentation Theory Obseration 4: h Aerae G h G G Since the reresentation reseres the rou structure we et: h Aerae G h G G

Reresentation Theory Obseration 4: h Aerae G h G G But this can be re-written as a summation oer the set hg: h Aerae G G hg

Reresentation Theory Obseration 4: h Aerae G G hg And since hg=g this imlies that: h Aerae G G G Aerae G

Reresentation Theory Obseration 5: Gien a inite rou G and ien a ector V the aerae o oer G is the closest G-inariant ector to : Aerae G ar min V G

Reresentation Theory Obseration 5: Aerae G ar min V G Since is inariant under the action o G we can write out the squared distances as: G G

Reresentation Theory Obseration 5: Since the reresentation is unitary we can rewrite this as: G G G G

Reresentation Theory Obseration 5: Since the reresentation reseres the rou structure we et: G G G G

Reresentation Theory Obseration 5: Re-writin this as a summation oer G - we et: G G G G

Reresentation Theory Obseration 5: And inally usin the act that the set G - is just a re-orderin o the set G we et: G G G G

Reresentation Theory Obseration 5: G G Thus is the G-inariant ector minimizin the squared distance to i and only i it minimizes the sum o squared distances to the ectors: n

Reresentation Theory Obseration 5: G G Thus is the G-inariant ector minimizin the squared distance to i and only i it minimizes the sum o squared distances to the ectors: n So must be the aerae o these ectors: Aerae G G G

Reresentation Theory Note: Since the aerae ma: Aerae G G is a linear ma returnin the closest G-inariant ector to the aerae ma is just the rojection ma rom V to V G : G G Aerae G

Outline Reresentation Theory Symmetry Detection Rotational Symmetry Relectie Symmetry

Symmetry Detection For unctions on a circle we deined measures o: Relectie Symmetry: or eery axis o relectie symmetry. Rotational Symmetry: or eery order o rotational symmetry. o 45 o Relectie 3-Fold Rotational 8-Fold Rotational

Symmetry Detection For unctions on a shere we would lie to deine a measure o: Relectie Symmetry: or eery lane o relectie symmetry.

Symmetry Detection For unctions on a shere we would lie to deine a measure o: Relectie Symmetry: or eery lane o relectie symmetry. Rotational Symmetry: or eery axis assin throuh the oriin and eery order o rotational symmetry.

Symmetry Detection Goal: Relectie Symmetry: Comute the sherical unction iin the measure o relectie symmetry about the lane erendicular to eery axis throuh the oriin. Rotational Symmetry: For eery order o rotational symmetry :» Comute the sherical unction iin the measure o -old symmetry about eery axis throuh the oriin.

Symmetry Detection Goal: Relectie Symmetries -Fold Rotational Symmetries Model 3-Fold Rotational Symmetries 4-Fold Rotational Symmetries

Symmetry Detection Aroach: As in the D case we will comute the symmetries o a shae by reresentin the shae by a sherical unction. Model Sherical Extent Function

Symmetry Detection Recall: To comute the measure o symmetry o a unction we: Associated a rou G o transormations to each tye o symmetry

Symmetry Detection Recall: To comute the measure o symmetry o a unction we: Associated a rou G o transormations to each tye o symmetry Deined the measure o symmetry as the size o the closest G-inariant unction: Sym G G

Symmetry Detection Recall: Usin the act that nearest symmetric unction was the aerae o the unction under the imae o the rou we ot: Sym G G G

Outline Reresentation Theory Symmetry Detection Rotational Symmetry Relectie Symmetry

Rotational Symmetry Gien a unction on the shere and ien a ixed order o rotational symmetry we would lie to deine a unction whose alue at eery oint oint is the measure o -old rotational symmetry about the associated axis. = =3 =5

Rotational Symmetry To do this we need to associate a rou to eery axis assin throuh the oriin.

Rotational Symmetry To do this we need to associate a rou to eery axis assin throuh the oriin. In articular i we denote by G the rou o - old rotations about the axis throuh : / y z x

Rotational Symmetry To do this we need to associate a rou to eery axis assin throuh the oriin. In articular i we denote by G the rou o - old rotations about the axis throuh the elements o the rou are the y rotations: j R corresondin to rotations about by the anle j /. x j z /

Rotational Symmetry Usin this notation the equation or the measure o -old symmetry o a unction about the axis becomes: Sym G j j

Rotational Symmetry Usin this notation the equation or the measure o -old symmetry o a unction about the axis becomes: Exandin this exression in terms o dotroducts we et: Sym j G j Sym j i G j i

Rotational Symmetry Usin this notation the equation or the measure o -old symmetry o a unction about the axis becomes: Usin the linearity o the inner-roduct we et: Sym j i G j i Sym j i G j i

Rotational Symmetry Usin this notation the equation or the measure o -old symmetry o a unction about the axis becomes: And usin the act that the reresentation is unitary we et: Sym j i G j i Sym j i G i j

Rotational Symmetry We can urther simliy this exression by obserin that the rotation j-i only deends on the dierence between the index j and i. Sym j j i G j i j

Rotational Symmetry We can urther simliy this exression by obserin that the rotation j-i only deends on the dierence between the index j and i. In articular usin the act that eery index in the rane [ can be exressed in exactly dierent ways as the dierence j-i with ji [ we et : Sym j j i G j i j

Rotational Symmetry Thus the measure o -old rotational symmetry about the axis can be comuted by tain the aerae o the dot-roducts o the unction with its rotations about the axis. Sym G j j

Rotational Symmetry Sym G j j So comutin the measures o rotational symmetry reduces to the roblem o comutin the correlation o with itsel: Dot R R

Rotational Symmetry Sym G j j So comutin the measures o rotational symmetry reduces to the roblem o comutin the correlation o with itsel: Dot R And this is somethin that we can do usin the Winer D-transorm rom last lecture. R

Rotational Symmetry Sym G j j Alorithm: Gien a unction : Comute the auto-correlation o. For each order o symmetry :» Comute the sherical unction whose alue at is the aerae o the correlation alues at rotations R j/ j.

Rotational Symmetry Sym G j j Comlexity: Comute the auto-correlation: On 3 lo n For each order o symmetry : Comute the sherical unction: On For a total comutational comlexity o On 4 to comute all rotational symmetries.

Rotational Symmetry Sym G Comlexity: Note: There is a lot o redundancy Comute the auto-correlation: On 3 lo n in this comutation. For each order o symmetry : Comute the sherical unction: On j j For a total comutational comlexity o On 4 to comute all rotational symmetries.

Rotational Symmetry Sym G Comlexity: Note: There is a lot o redundancy Comute the auto-correlation: On 3 lo n in this comutation. For Examle: each order Comutin o symmetry the -old : symmetries Comute the sherical unction: On we re-comute the -old symmetry allowin or a -old imroement in eiciency For a total comutational comlexity o On 4 to comute all rotational symmetries. j j

Outline Reresentation Theory Symmetry Detection Rotational Symmetry Relectie Symmetry

Relectie Symmetry Gien a sherical unction we would lie to comute a unction whose alue at a oint is the measure o relectie symmetry with resect to the lane erendicular to.

Relectie Symmetry Relections throuh the lane erendicular to corresond to a rou with two elements: G Id Re

Relectie Symmetry Relections throuh the lane erendicular to corresond to a rou with two elements: G Id Re So the measure o relectie symmetry becomes: Sym G Re Re

Relectie Symmetry Relections throuh the lane erendicular to corresond to a rou with two elements: G Id Re So the measure o relectie symmetry becomes: Sym G Re Re

Relectie Symmetry How do we comute the dot-roduct o the unction with the relection o throuh the lane erendicular to?

Relectie Symmetry How do we comute the dot-roduct o the unction with the relection o throuh the lane erendicular to? Since relections are not rotations they hae determinant - we cannot use the alues o the autocorrelation.

Relectie Symmetry General Aroach: I we hae two relections S and T we can set R to be the transormation: R TS

Relectie Symmetry General Aroach: I we hae two relections S and T we can set R to be the transormation: R TS Since S and T are both orthoonal the roduct R must also be orthoonal.

Relectie Symmetry General Aroach: I we hae two relections S and T we can set R to be the transormation: R TS Since S and T are both orthoonal the roduct R must also be orthoonal. Since both S and T hae determinant - R must hae determinant.

Relectie Symmetry General Aroach: I we hae two relections S and T we can set R to be the transormation: R Thus R must be a rotation and we hae: RS TS so T is just the relection S ollowed by some rotation. T

Relectie Symmetry General Aroach: Thus i we comute the correlation o with some relection S : Dot R S We can obtain the dot roduct o with its relection throuh the lane erendicular to by ealuatin: Re Dot S R Re S Rotation S

Relectie Symmetry Sym Alorithm: Gien a unction : G Dot Re Comute the correlation o with the relection S Comute the sherical unction whose alue at is the aerae o the size o and the dot-roduct o with the rotation o S by Re S -. S S

Relectie Symmetry Sym Comlexity: G Dot Re Comute the correlation: On 3 lo n Comute the sherical unction: On S S For a total comutational comlexity o On 3 lo n to comute all relectie symmetries.

Relectie Symmetry Sym Comlexity: G Dot Re Comute the correlation: On 3 lo n For comutin relectie symmetries the comutation o the correlation is oerill as we don t use most o the correlation alues. Comute the sherical unction: On For a total comutational comlexity o On 3 lo n to comute all relectie symmetries. S S

Relectie Symmetry There are many dierent choices or the relection S we use to comute: Dot S T

Relectie Symmetry There are many dierent choices or the relection S we use to comute: Dot T The simlest relection we can use is the antiodal ma: S S

Relectie Symmetry The adantae o usin the antiodal ma is that it maes it easy to exress Re S -.

Relectie Symmetry The adantae o usin the antiodal ma is that it maes it easy to exress Re S -. In articular ixin a oint we can thin o S as the combination o two mas: A relection throuh the lane erendicular to and A rotation by 8 o about the axis throuh.

Relectie Symmetry Thus a relection throuh the lane erendicular to can be exressed as the roduct o the antiodal ma and a rotation by 8 o around the axis throuh : Re R S

Relectie Symmetry Thus settin S to be the antiodal ma we et: Sym G R S

Relectie Symmetry Usin the sherical harmonics we et a simle exression or S :

Relectie Symmetry Y Im Y Y Re Y Im Y Im Y Y Re Y Re Y Im Y 3 3 Im Y 3 Im Y 3 Y 3 Re Y Re 3 Y 3 Re 3 Y 3

Relectie Symmetry Usin the sherical harmonics we et a simle exression or S : S l l ˆ l m l m l Y m l

Relectie Symmetry Additionally i the unction is antiodally symmetric: the equation or the relectie symmetries becomes: S Sym Sym R S R G G

Relectie Symmetry Additionally i the unction is antiodally symmetric: the equation or the relectie symmetries becomes: S Sym Sym R S R G G

Relectie Symmetry Additionally i the unction is antiodally symmetric: the equation or the relectie symmetries becomes: S Sym Sym R S R G G

Relectie Symmetry That is in the case that is antiodally symmetric the relectie symmetries o and the -old rotational symmetries o are equal. -Fold Symmetry Benzene Relectie Symmetry