1 Scale and Rotation Invariant Matching Using Linearly Augmented Tree Hao Jiang Boston College Tai-peng Tian, Stan Sclaroff Boston University
Scale and Rotation Invariant Matching 2
Previous Methods Hough Transform (Duda & Hart) and RANSAC (Fischler and Bolles) Dynamic programming (Felzenszwalb and Huttenlocher 05) Loopy belief propagation (Weiss and Freeman 01) Tree-reweighted message passing (Kolmogorov 06) Primal-dual methods (Komodakis and Tziritas 07) Dual decomposition (Komodakis, Paragios and Tziritas 11, Torresani, Kolmogorov and Rother 08) Successive convexification (Jiang 2009, Li, Kim, Huang and He 2010) 3
Unsolved Issue How to find the optimal rotation angle and scale especially if the ranges are unknown? 4 Quantizing rotation angle and scale
The Model We all Want to Have 5
6 In Reality We Need to Use … Hyperedge Non-tree edge
Linearly Augmented Tree (LAT) 7 Any tree constraints Linear non-tree constraints LAT works on continuous scale and rotation and non-tree structure.
Optimizing Invariant Matching 8 Total local feature matching cost p f(p) q f(q) cost(p,f(p)) cost(q,f(q)) … …
9 In matrix form: X X Binary assignment Matrix C C Local matching cost matrix Optimizing Invariant Matching
10 Rotation and scaling consistency Model tree edges Target Optimizing Invariant Matching
11 Y p,q Pairwise assignment matrix for site pair (p,q) s 0, u 0 = sin(θ 0 ), v 0 = cos(θ 0 ) Θ p,q S p,q Rotation angle matrixScale matrix Optimizing Invariant Matching
12 Other linear global terms such as area constraints or global affine constraints. Optimizing Invariant Matching Area scaling is
The Mixed Integer Optimization 13 g(X) Subject To: Constraints on binary matrices X, Y, and continuous variables u0, v0 and s0. Unary matching cost Rotation consistency Scaling term Other global terms
Linear Relaxation 14
Special Structure 15 Objective function “Hard” constraints X, Y Auxiliary variables Auxiliary variables Easy Ones
The Solution Space 16 Solutions feasible to the “simple” problem. Solutions feasible to “hard” constraints. Optimal solution
17 Column Generation Two initial proposals and the current best estimate. Proposals
18 Column Generation Proposals Few extreme points (proposals) can be used to obtain the solution, and they can be generated iteratively.
Decompose into Dynamic Programming 19 Create the initial trellises, and find first 2 proposals k=2 Create the initial trellises, and find first 2 proposals k=2 Find out how to update the tree Update the trees Dynamic Programming and generate new proposal (k+1) Dynamic Programming and generate new proposal (k+1) Gain > 0 Yes Done
An Example 20 Template Image
An Example 21 Template
An Example 22 Target Image
An Example 23
Another Example 24 Template
Another Example 25 Target Image
Another Example 26
Complexity Comparison 27 Direct solution Direct solution
SIFT Matching Results 28 Detection Rate This paper DPRANSACTensor (Duchenne 09) Linear (Jiang 09) Affine (Li 00) 90%63%48%5%88%8%
Match Unreliable Regions 29 Detection Rate This paper DPRANSACTensor (Duchenne 09) Linear (Jiang 09) Affine (Li 00) 90%86%20%16%34%1%
Matching Unreliable Regions 30 Detection Rate This paper DPRANSACTensor (Duchenne 09) Linear (Jiang 09) Affine (Li 00) 91%73%43%11%74%14%
More Tests 31 RateTim e 90%0.78s 91%0.42s 98%0.03s 90%0.02s 94%0.07s 91%0.05s
Test on Ground Truth Data 32
Summary LAT method can incorporate global constraints and can be efficiently solved. It works even with very weak features and large deformations in scale and rotation invariant matching. It works on continuous scale and rotation and does not need an upper bound for the scale. The decomposition framework would be useful to enhance the widely applied tree methods in object detection, pose estimation and etc. 33
34 The End