Image analysis – Applications – 3-d or stereo imaging analysis
Reexamination Certificate
2001-04-03
2004-12-21
Couso, Yon J. (Department: 2625)
Image analysis
Applications
3-d or stereo imaging analysis
C382S201000, C382S218000, C356S419000
Reexamination Certificate
active
06834119
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
This invention generally relates to the field of image processing systems and methods, and more particularly relates to methods of matching image features across multiple image views.
2. Description of Related Art
There are many image processing applications where multiple images of scenes are matched to identify common image information across the different images. For example, one way of creating three-dimensional (3-D) digital content is by analyzing multiple images of a scene. The main issue here is image matching—how to automatically find the corresponding relationship among the projections in different images of the same points.
For example, as shown in
FIG. 1
, a left eye
102
is seen in three images
104
,
106
,
108
, of the same scene
110
. The left eye
102
appears as three respective image features
112
,
114
,
116
, in the three images
104
,
106
, and
108
, as shown. Image matching methods attempt to establish links
118
,
120
among the three respective image features
112
,
114
,
116
, that are common across the three images
104
,
106
,
108
, of the same scene
110
. The significance of image matching is that once a correspondence information is established, it is possible to recover the 3-D coordinates of the matched points from which more complete geometric structures can be reconstructed.
Two prior art methods have been reported which work on image pairs or triplets, respectively. For example, see the publication by R. Deriche, Z. Zhang, Q.-T. Luong and O. Faugeras, “Robust Recovery of the Epipolar Geometry for an Uncalibrated Stereo Rig,” Proc. European Conference on Computer Vision '94, pp. 567-576. Additionally, see the publication by P. H. S. Torr and A. Zisserman, “Robust Parameterization and Computation of the Trifocal Tensor,” Image and Vision Computing, Vol. 15, No. 8, August 1997, pp. 591-605.
The basic approach is, first, to generate a number of candidate correspondences based on proximity and similarity; and then, to select the correct ones from all candidates by making sure that they satisfy an algebraic constraint (epipolar geometry in the two-view case, and trifocal tensor in the three-view case). In the terminology of estimation theory, the correct candidates are called inliers, whilst the wrong ones are called outliers. The robustness of a method is its ability to detect outliers. Unfortunately, the robustness of the two prior art methods mentioned above is limited because the constraints they are enforcing are ambiguous sometimes. That is, there may be multiple pairs or triplets of correspondences that satisfy the same instance of a constraint. Additionally, those constraints have singular conditions, e.g. when the camera positions are linear or planar. Under such cases, these two methods simply fail to work.
Therefore a need exists to overcome the problems with the prior art as discussed above, and particularly for a method and apparatus that can more successfully match features across multiple images.
SUMMARY OF THE INVENTION
According to a preferred embodiment of the present invention, an image processing system comprises a memory; a controller/processor electrically coupled to the memory; an image feature detector, electrically coupled to the controller/processor and to the memory, for detecting a plurality of image features in a first image corresponding to a first view of a scene, and for detecting a plurality of image features in at least a second image corresponding to a respective at least a second view of the scene, wherein the at least a second image deviates from the first image as a result of camera relative motion; and an image matching module, electrically coupled to the controller/processor and to the memory, for determining a two-view correspondence resulting in a potential match set of candidate image features between the first image and the at least a second image, wherein the potential match set is determined to have a maximum average strength of correspondence based at least in part on the total number of matching neighbor candidate image features for each match of the potential match set.
According to a preferred embodiment of the present invention, an image processing system comprises a memory; a controller/processor electrically coupled to the memory; an image feature detector, electrically coupled to the controller/processor and to the memory, for detecting a plurality of image features in a first image corresponding to a first view of a scene, and for detecting a plurality of image features in at least a second image corresponding to a respective at least a second view of the scene, wherein the at least a second image deviates from the first image as a result of camera relative motion; and an image matching module, electrically coupled to the controller/processor and to the memory, for determining a multiple-view correspondence between the plurality of detected features in the first image and the plurality of detected image features in the at least a second image, resulting in a potential match set of candidate image features between the first image and the at least a second image, wherein the potential match set is based at least in part on a computation of reprojection error for matched points that resulted from a projective reconstruction of the potential match set.
REFERENCES:
patent: 4982438 (1991-01-01), Usami et al.
patent: 5475422 (1995-12-01), Mori et al.
patent: 5675377 (1997-10-01), Gibas
patent: 5850463 (1998-12-01), Horii
patent: 6665440 (2003-12-01), Zhang et al.
Deriche, R., et al., “Robust Recovery of the Epipolar Geometry for an Uncalibrated Stereo Rig,” Proc. European Conference on Computer Vision, 1994, pp. 567-576.
Torr, P.H.S., et al., “Robust Parameterization and Computation of the Trifocal Tensor,” Image and Vision Computing, vol. 15, No. 8, Aug. 1997, pp. 591-605.
Chen, Q., et al., “Efficient Iterative Solutions to M-View Projective Reconstruction Problem,” Proc. Computer Vision and Pattern Recognition 1999, vol. II, pp. 55-61.
Bongini Stephen C.
Chawan Sheela
Couso Yon J.
Jorgenson Lisa K.
STMicroelectronics Inc.
LandOfFree
Methods and apparatus for matching multiple images does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Methods and apparatus for matching multiple images, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Methods and apparatus for matching multiple images will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3277595