Automatic Registration of 2D and 3D Images

1
Automatic Registration of 2-D and 3-D Images

• Yuanxin Zhu
• Multimedia Communication Visualization Lab,
• University of Missouri-Columbia

2
Outline

• 1. Introduction
• 2. 2-D Image registration
• 3. Automatic registration of partially
  overlapping range images

3
References

• Zheng93 Q Zheng and R Chellappa, "A
  Computational Vision Approach to Image
  Registration," IEEE Transactions on Image
  Processing, vol. 2, No. 3, pp. 313-326, July
  1993.
• Hsieh97 J-W Hsieh, H-Y Liao, K-C Fan, et al.,
  "Image Registration Using a New Edge-Based
  Approach," Computer Vision and Image
  Understanding, vol. 67, no. 2, August, pp.
  112-130, 1997.
• Chen97 C-S Chen, Y-P Hung, C-C Chiang, et al.,
  "Rang Data Acquisition Using Color Structured
  Lighting and Stereo Vision," Image and Vision
  Computing, vol. 15, pp. 445-456, 1997.
• Chen99 C-S Chen, Y-P Hung, J-B Cheng,
  "RANSAC-based DARCES A New Approach to Fast
  Automatic Registration of Partially Overlapping
  Range Images," Proceedings of ICCV98.

4
1. Introduction

• Image registration is an important technique for
  a great variety of applications such as
• Aerial image analysis
• Stereo vision
• Automated cartography
• Motion analysis
• The recovery of the 3D characteristics of a
  scene.
• Summarizes the methods proposed in the 4
  references which address issues related to 2-D
  Image registration, range image acquisition, and
  its automatic registration.

5
2. 2-D Image Registration

• 2.1 The 2-D image registration problem
• In 2-D image registration, there are often two
  assumptions
• Images are taken by cameras whose optical axes
  are parallel
• The variations of the intensity characteristics
  between images are assumed to be small.
• Under these assumptions, 2-D image registration
  can be described as finding out a set of
  parameters that completes the affine transform
  between the two frames as shown in the following
  equation.

6
2.2 2-D Image Registration Methods

• Two tasks are needed to be handled during an
  image registration process
• feature selection
• feature correspondence establishment.
• Algorithms for determining feature
  correspondence can be classified into two
  categories
• feature-based (curvatures, moments, areas, line
  segments)
• area-based methods (usually adopts a window of
  points to determine a matched location using the
  correlation technique Brown92)

7
2.2 2-D Image Registration Methods

• Area-based method is more robust than the
  feature-based method in some situations. However,
  if the orientation difference between the two
  images is large, the value of cross-correlation
  will be greatly influenced and the
  correspondences between feature points, thus,
  hard to derive.
• In order to solve the problem, it is necessary to
  develop a method to estimate the rotation
  parameter in advance.

8
2.3 Zheng and Chellappa Method

• In Zheng and Chellappa's approach, the technique
  for estimating the rotation angle works well for
  most cases.
• However, if a scene includes many buildings and
  objects, the method will fail due to the fact
  that the illumination conditions in one image may
  not be equivalent to those in the other.
• In general, the estimation of a rotation angle in
  their approach is rough. Further, their approach
  requiring a Gabor function decomposition is
  computationally intensive.
• Another drawback of their approach is that when
  false matches emerge, their method cannot handle
  them.

9
2.4 Hsieh et al. Method

• Hsieh et al. Hsieh97 proposes an area-based
  two-stage scheme to tackle the above-mentioned
  problems. The purpose of the first stage is to
  obtain the initial values of the registration
  parameters. In the second stage, an iterative
  procedure is proposed to repeatedly refine the
  registration parameters.
• In the first stage, the scheme applies wavelet
  transform to extract a number of feature points
  as the basis for registration.
• By using a line-fitting model, all the edge
  directions of the edge points are estimated from
  the edge outputs of a transformed image.

10
2.3 Hsieh et al. Method (Continue)

• In order to estimate the orientation
  difference between the images, a so-called "angle
  histogram" is calculated. From the angle
  histogram, the rotation angle which can be used
  to compensate for the difference between the two
  target images can be decided by seeking the angle
  that corresponds to the maximum peak in the
  histogram.
• Based on the rotation angle, an initial
  matching can be performed.
• During the second stage, they apply an
  iterative scheme to make the registration result
  more reliable.

11
3. Automatic Registration of Partially
Overlapping Range Images

• 3.1 Definition of 3D registration problem

12
3.2 Data-Aligned Rigidity-Constrained Exhaustive
Search (DARCES)
13
3.3 RANSAC-Based DARCES Approach for
Partially-Overlapping Case