*McKee, J.W.*, and
*Aggarwal, J.K.*,

**Computer Recognition of Partial Views of Curved Objects**,

*TC(26)*, No. 8, August, 1977, pp. 790-800.
BibRef
**7708**

Earlier:

**Computer Recognition of Partial Views of Three Dimensional
Curved Objects**,

*ICPR76*(499-503).
BibRef

And:
*Univ. of Texas*CS TR 171, 1975.
*Recognize Two-Dimensional Objects*. A Curve is represented by a variation of the chain code and
matching is performed using this representation.
BibRef

*Perkins, W.A.*,

**A Model-Based Vision System for Industrial Parts**,

*TC(27)*, No. 2, February 1978, pp. 126-143.
*Industrial Applications*.
BibRef
**7802**

Earlier:

**A Model Based Vision System for Scenes Containing Multiple Parts**,

*IJCAI77*(678-684).
Represent curves using orientation vs. length and "correlate" boundary
fragments in this representatio.
See also INSPECTOR: A Computer Vision System That Learns to Inspect Parts.
BibRef

*Perkins, W.A.*,

**Model-Based Inspection System for Component Boards**,

*PR(17)*, No. 1, 1984, pp. 135-140.

Elsevier DOI
BibRef
**8400**

Earlier:

**Simplified Model-Based part Locator**,

*ICPR80*(260-263).
Correct component, correct position.
BibRef

*Davis, L.S.*,

**Shape Matching Using Relaxation Techniques**,

*PAMI(1)*, No. 1, January 1979, pp. 60-72.
BibRef
**7901**

Earlier:
*PRIP77*(191-197).
*Matching, Contours*.
*Relaxation*. Match outlines of islands extracted from a map.
This program used a spring template description of shapes (outlines or
portions of outlines) and a relaxation based search procedure to find
the best match. Complete boundaries from a geographic data-base were
used to generate the models. Test templates were composed of portions
of similar outlines. Both the model and template were formed from a
line segment representation of the contour using a given threshold on
the curvature of the underlying boundary curve. These curves are then
represented by the sequence of angles and the segments between
adjacent angles. Each angle forms a sub-template with pairs of these
connected by springs. A local match of a pair of template elements
(angles) with a pair of model elements results in tension in the
spring joining the template elements. The goal is to find the
assignment that minimizes the tension in the springs. To find this
minimum, a discrete relaxation procedure is used. First an
association graph is constructed where a node corresponds to a single
template elements to model element pairing with a weight determined by
the local evaluation function (spring tension). The links between
nodes correspond to pairs of assignments and are included only if the
spring tension caused by the assignments is low enough. (The
magnitude of the threshold controls how many possible assignments are
tried.) Nodes are pruned if the total neighborhood tension is too
large. A second edge (arc) filtering procedure removes the links if
the global transform generated from the match for each of the nodes
are different. Each of these discrete relaxation operations differs
from other discrete methods in that links or nodes only must be
adequately consistent, not absolutely consistent.
These steps produce a pruned association graph with (possibly) several
disjoint subgraphs. Each subgraph corresponds to one global transform
to align the template with the model object. The best match is
determined by evaluating the several final matches using the spring
template evaluation function. This matching procedure is invariant to
rotation and translation since the angle between segments is the only
direction used. The use of approximate consistency allows for small
scale changes or distortions.
BibRef

*Davis, L.S.*,
*Rosenfeld, A.*,

**An Application of Relaxation Labelling to Spring-Loaded Template
Matching**,

*ICPR76*(591-597).
BibRef
**7600**

*Henderson, T.C.[Thomas C.]*,
*Davis, L.S.[Larry S.]*,

**Hierarchical Models and Analysis of Shape**,

*PR(14)*, No. 1-6, 1981, pp. 197-204.

Elsevier DOI
BibRef
**8100**

*Davis, L.S.[Larry S.]*,
*Henderson, T.C.[Thomas C.]*,

**Hierarchical Constraint Processes for Shape Analysis**,

*PAMI(3)*, No. 3, 1981, pp. 265-277.
BibRef
**8100**

*Davis, L.S.*,

**Representation and Recognition of Cartographic Data**,

*MDP80*(169-189).
BibRef
**8000**

*Hakalahti, H.*,
*Harwood, D.*,
*Davis, L.S.*,

**Two-Dimensional Object Recognition by Matching Local Properties
of Contour Points**,

*PRL(2)*, 1984, pp. 227-234.
BibRef
**8400**

*Ayache, N.J.[Nicholas J.]*, and
*Faugeras, O.D.*,

**HYPER: A New Approach for the Recognition and Positioning of
Two-Dimensional Objects**,

*PAMI(8)*, No. 1, January 1986, pp. 44-54.
BibRef
**8601**

Earlier:

**A New Method for the Recognition and Positioning of 2-D Objects**,

*ICPR84*(1274-1277).
BibRef

Earlier: A1 only:

**A Model Based Vision System to Identify and Locate Partially
Visible Industrial Parts**,

*CVPR83*(492-494).
*Matching, Contours*.
*Recognize Two-Dimensional Objects*. The model and the image are composed of closed contours of
industrial parts. These parts can have extra pieces (sprues) and
can be overlapping, thus there is a need to match when there are
few actual matching segments. The procedure works by first finding a few
matches using privileged segments (usually long segments).
These starting matches
are used to generate a transformation from model to image.
The hypothesis is evaluated by finding other segments that
match using position along the contour.
There are some good things, but it can be made
simpler.
BibRef

*Faugeras, O.D.*,
*Ayache, N.J.*,
*Faverjon, B.*,

**A Geometric Matcher for Recognizing and Positioning 3-D Rigid Objects**,

*CAIA84*(218-224).
BibRef
**8400**

*Ayache, N.J.*,
*Faverjon, B.*,
*Boissonnat, J.D.*,
*Bollack, B.*,

**Automatic Handling of Overlapping Workpieces**,

*ICPR84*(837-839).
BibRef
**8400**

*Avis, D.*,
*Elgindy, H.*,

**A Combinatorial Approach to Polygon Similarity**,

*IT(29)*, 1983, pp. 148-150.
BibRef
**8300**

*Berman, S.*,
*Parikh, P.*,
*Lee, C.S.G.*,

**Computer Recognition of Two Overlapping Parts Using a Single Camera**,

*Computer(18)*, No. 3, March 1985, pp. 70-80.
BibRef
**8503**

*Turney, J.L.*,
*Mudge, T.N.*, and
*Volz, R.A.*,

**Recognizing Partially Occluded Parts**,

*PAMI(7)*, No. 4, July 1985, pp. 410-421.
(Univ. of Mich.)
*Recognize Two-Dimensional Objects*. Not really contour matching, but they use binary templates
represented as boundaries. The first step is to determine the
salient features and weight these more strongly in the match, a form
of template matching.
BibRef
**8507**

*Vernon, D.[David]*,

**Two-Dimensional Object Recognition Using Partial Contours**,

*IVC(5)*, No. 1, February 1987, pp. 21-27.

Elsevier DOI
BibRef
**8702**

*Wallace, A.M.*,
*Brodie, E.E.*,
*Love, S.*,

**Discrimination Between Visual Stimuli by Variation of Shape
and Relative Position of Volumetric Primitives**,

*IVC(11)*, No. 6, July-August 1993, pp. 372-388.

Elsevier DOI Set of global properties, a set of shape descriptions of the
individual components, and a description of pairwise relations between
components.
BibRef
**9307**

*Wallace, A.M.[Andrew M.]*,

**An Informed Strategy for Matching Models to Images of
Fabricated Objects**,

*PR(20)*, No. 3, 1987, pp. 349-363.

Elsevier DOI Matching 2D scene to models.
BibRef
**8700**

*Wallace, A.M.[Andrew M.]*,

**Matching Segmented Scenes to Models Using Pairwise Relationships
between Features**,

*IVC(5)*, No. 2, May 1987, pp. 114-120.

Elsevier DOI
BibRef
**8705**

*Paglieroni, D.W.[David W.]*,
*Jain, A.K.[Anil K.]*,

**Fast Classification Of Discrete Shape Contours**,

*PR(20)*, No. 6, 1987, pp. 583-598.

Elsevier DOI
BibRef
**8700**

*Manay, S.[Siddharth]*,
*Paglieroni, D.W.[David W.]*,

**Matching Flexible Polygons to Fields of Corners Extracted from Images**,

*ICIAR07*(447-459).

Springer DOI
**0708**

BibRef

*Gorman, J.W.*,
*Mitchell, O.R.*, and
*Kuhl, F.P.*,

**Partial Shape Recognition Using Dynamic Programming**,

*PAMI(10)*, No. 2, March 1988, pp. 257-266.

IEEE DOI Yet another contour recognition experiment.
BibRef
**8803**

*Kamgar-Parsi, B.[Behzad]*,
*Margalit, A.[Avraham]*,
*Rosenfeld, A.[Azriel]*,

**Matching General Polygonal Arcs**,

*CVGIP(53)*, No. 2, March 1991, pp. 227-234.

Elsevier DOI
BibRef
**9103**

And:
Erratum:
*CVGIP(54)*, No. 2, September 1991, pp. 307.

Elsevier DOI Break into equal length line segments and match using sum of
squared distances between corresponding points on the arcs. Use
for finding the piece of a long arc that matches a short arc.
BibRef

*Rao, N.S.V.*,
*Wu, W.*,
*Glover, C.W.*,

**Algorithms for Recognizing Planar Polygonal Configurations
Using Perspective Images**,

*RA(8)*, No. 4, August 1992, pp. 480-485.
Assume it is planar and matching is simplified.
BibRef
**9208**

*Ueda, N.*, and
*Suzuki, S.*,

**Learning Visual Models from Shape
Contours Using Multiscale Convex/Concave Structure Matching**,

*PAMI(15)*, No. 4, April 1993, pp. 337-352.

IEEE DOI
BibRef
**9304**

Earlier:

**Automatic Shape Model Acquisition Using Multiscale Segment Matching**,

*ICPR90*(I: 897-902).

IEEE DOI Extract the salient features from a set of
samples to produce a prototype of the object.
BibRef

*Milios, E.E.[Evangelos E.]*,

**Shape Matching Using Curvature Processes**,

*CVGIP(47)*, No. 2, August 1989, pp. 203-226.

Elsevier DOI Matching with some deformation. Represent the shape as concave and
convex segments of the contour, match segments using dynamic
programming, recover the differences.
BibRef
**8908**

*Mitiche, A.[Amar]*, and
*Aggarwal, J.K.*, (UTexas),

**Contour Registration by Shape-Specific Points for
Shape Matching**,

*CVGIP(22)*, No. 3, June 1983, pp. 396-408.

Elsevier DOI Registration is used as a preprocessor for contour shape matching.
The translation, rotation and scale parameters are computed by
assuming certain points are equivalent, e.g. centroid, points
farthest (nearest) from centroid, etc.
BibRef
**8306**

*della Ventura, A.*,
*Ongaro, P.*, and
*Schettini, R.*,

**Search and Replace of 2-D Objects in Digital Images**,

*VF91*(205-212).
Mostly an application of matching.
BibRef
**9100**

*Lu, C.C.*,
*Dunham, J.G.*,

**Shape-Matching Using Polygon Approximation and Dynamic Alignment**,

*PRL(14)*, No. 12, December 1993, pp. 945-949.
BibRef
**9312**

*Boissonnat, J.D.*,

**Stable Matching Between a Hand Structure and an Object Silhouette**,

*PAMI(4)*, No. 6, November 1982, pp. 603-612.
First compute the possible stable positions (for 1 finger) for all
possible local shapes (restricted set) and the conditions for a
valid grasp for the given gripper configuration. Experiments on
the bin of parts problem using range data. Good.
BibRef
**8211**

*Kashyap, R.L.*, and
*Oomman, B.J.*,

**A Geometrical Approach to Polygonal Dissimilarity and Shape Matching**,

*PAMI(4)*, No. 6, November 1982, pp. 649-654.
BibRef
**8211**

And:
*ICPR82*(472-479).
Two metrics, first overlap error, second minimum integral error
between polygons. Align 2 edges (at midpoints) go around the figure
and compute the difference between the corresponding points in each
(each step is the same proportion of the border length).
BibRef

*Smith, S.P.[Stephen P.]*,
*Jain, A.K.[Anil K.]*,

**Chord Distributions for Shape Matching**,

*CGIP(20)*, No. 3, November 1982, pp. 259-271.

Elsevier DOI
BibRef
**8211**

Earlier:
*PRIP81*(168-170).
BibRef

*You, Z.S.[Zhi-Sheng]*,
*Jain, A.K.[Anil K.]*,

**Performance Evaluation of Shape Matching via Chord Length Distribution**,

*CVGIP(28)*, No. 2, November 1984, pp. 185-198.

Elsevier DOI
*Matching, Evaluation*. (Michigan State) This paper uses outlines like those used by Davis
(above) but does not refer to his paper. The outlines were distorted
and then these results were used in the matching experiments.
BibRef
**8411**

*Alt, H.*,
*Behrends, B.*,
*Blomer, J.*,

**Approximate Matching of Polygonal Shapes**,

*AMAI(13)*, No. 3-4, 1995, pp. 251-265.
BibRef
**9500**

*Ventura, J.A.*,
*Nain, L.Y.*, and
*Wan, W.*,

**Optimal Matching of General Polygons Based on the Minimum Zone Error**,

*PRL(16)*, 1995, pp. 1125-1136.
BibRef
**9500**

*Ventura, J.A.*,
*Chen, J.M.*,

**Optimal Matching of Nonconvex Polygons**,

*PRL(14)*, 1993, pp. 445-452.
BibRef
**9300**

*Cox, P.*,
*Maitre, H.*,
*Minoux, M.*,
*Ribeiro, C.C.[Celso C.]*,

**Optimal Matching of Convex Polygons**,

*PRL(9)*, 1989, pp. 327-334.
BibRef
**8900**

*Griffin, P.M.*,

**Correspondence of 2-D Projections by Bipartite Matching**,

*PRL(9)*, 1989, pp. 361-366.
BibRef
**8900**

*Kamgar-Parsi, B.[Behzad]*,
*Kamgar-Parsi, B.[Behrooz]*,

**Matching Sets of 3D Line Segments with Application to
Polygonal Arc Matching**,

*PAMI(19)*, No. 10, October 1997, pp. 1090-1099.

IEEE DOI
**9710**

BibRef

Earlier:

**Matching 3-D Arcs**,

*CVPR97*(28-33).

IEEE DOI
**9704**

Equal length line segments. Representation of arcs. Find shortest arc in
long curve.
Decompose the contour and match.
BibRef

*Kamgar-Parsi, B.[Behzad]*,
*Kamgar-Parsi, B.[Behrooz]*,

**Algorithms for Matching 3D Line Sets**,

*PAMI(26)*, No. 5, May 2004, pp. 582-593.

IEEE Abstract.
**0404**

Solution of the Finite-to-Finite, Finite-to-Infinite and Infinite-to-Infinite
matching problems.
Built on
See also Matching of 3D Polygonal Arcs. and
See also Matching Sets of 3D Line Segments with Application to Polygonal Arc Matching.
BibRef

*Kamgar-Parsi, B.*,
*Kangar-Parsi, B.*,

**An invariant, closed-form solution for matching sets of 3D lines**,

*CVPR04*(II: 431-436).

IEEE DOI
**0408**

BibRef

*Kamgar-Parsi, B.[Behzad]*,
*Kamgar-Parsi, B.[Behrooz]*,

**An Open Problem in Matching Sets of 3D Lines**,

*CVPR01*(I:651-656).

IEEE DOI
**0110**

BibRef

And:

**Line Matching: Solutions and Unsolved Problems**,

*ICIP01*(II: 905-908).

IEEE DOI
**0108**

BibRef

*Chen, J.M.*,

**Optimal Matching of Closed Contours with Line Segments and Arcs**,

*PRL(18)*, No. 6, June 1997, pp. 567-574.
**9710**

BibRef

*Shan, Y.[Ying]*,
*Zhang, Z.Y.[Zheng-You]*,

**New Measurements and Corner-Guidance for Curve Matching with
Probabilistic Relaxation**,

*IJCV(46)*, No. 2, February 2002, pp. 157-171.

DOI Link
**0201**

BibRef

Earlier: A2, A1:

**A Progressive Scheme for Stereo Matching**,

*SMILE00*(68 ff.).

Springer DOI
**0209**

BibRef

*Zabulis, X.[Xenophon]*,
*Sporring, J.[Jon]*,
*Orphanoudakis, S.C.[Stelios C.]*,

**Perceptually relevant and piecewise linear matching of silhouettes**,

*PR(38)*, No. 1, January 2005, pp. 75-93.

Elsevier DOI
**0410**

Correspondences of landmarks then the boundaries between landmarks.
BibRef

*Liu, H.R.[Hai-Rong]*,
*Cao, S.J.[Sheng-Jiao]*,
*Yan, S.C.[Shui-Cheng]*,

**Automated Assembly of Shredded Pieces From Multiple Photos**,

*MultMed(13)*, No. 5, 2011, pp. 1154-1162.

IEEE DOI
**1110**

BibRef

*Ataer-Cansizoglu, E.[Esra]*,
*Bas, E.[Erhan]*,
*Kalpathy-Cramer, J.[Jayashree]*,
*Sharp, G.C.[Greg C.]*,
*Erdogmus, D.[Deniz]*,

**Contour-based shape representation using principal curves**,

*PR(46)*, No. 4, April 2013, pp. 1140-1150.

Elsevier DOI
**1301**

Shape representation and analysis; Curve/contour matching
BibRef

*Laiche, N.[Nacéra]*,
*Larabi, S.[Slimane]*,
*Ladraa, F.[Farouk]*,
*Khadraoui, A.[Abdelnour]*,

**Curve normalization for shape retrieval**,

*SP:IC(29)*, No. 4, 2014, pp. 556-571.

Elsevier DOI
**1404**

Curvature points
BibRef

HTML Version.

BibRef

*Richardson, T.*,
*Wang, S.[Song]*,

**Nonrigid Shape Correspondence Using Landmark Sliding,
Insertion and Deletion**,

*MICCAI05*(II: 435-442).
BibRef
**0500**

*Wang, S.[Song]*,
*Kubota, T.*,
*Richardson, T.*,

**Shape correspondence through landmark sliding**,

*CVPR04*(I: 143-150).

IEEE DOI
**0408**

Match a set of landmarks along the contour.
BibRef

*Rothwell, C.A.*,

**Reasoning about Occlusions During Hypothesis Verification**,

*ECCV96*(I:599-609).

Springer DOI Analysis of matching methods.
BibRef
**9600**

*Serra, B.[Bruno]*,
*Berthod, M.[Marc]*,

**Optimal Subpixel Matching of Contour Chains and Segments**,

*ICCV95*(402-407).

IEEE DOI
BibRef
**9500**

*Serra, B.*,
*Berthod, M.*,

**Subpixel Contour Matching Using Continuous Dynamic Programming**,

*CVPR94*(202-207).

IEEE DOI
BibRef
**9400**

*Price, K.E.*,

**Matching Closed Contours**,

*CVWS84*(130-134).
BibRef
**8400**
*USC Computer Vision*
BibRef

And:
*DARPA84*(169-175).
BibRef

And:
*ICPR84*(990-992).
Closed contours provide many constraints on the possible matches.
This paper uses a simple idea to match contours of single objects with
collections of these objects that include many occlusions and
overlaps. Initial matches are found using only the angle between
segments as the feature. Several (3) consecutive matching angles
indicate a possible match and give a transformation (rotation and
translation) that would align the contour model contour with at least
part of the image contour. These possible matches are checked by
transforming the model contour and checking for overlaps of the line
segments representing the contour in a manner much like the work of
Clark(
See also Matching of Natural Terrain Scenes. ) or Medioni (
See also Matching Images Using Linear Features. ).
BibRef

*Zielke, T.*,
*von Seelen, W.*,

**Matching Conic Curve Segments**,

*ICPR92*(I:583-586).

IEEE DOI
BibRef
**9200**

*Jacobs, D.W.*,

**GROPER: A Grouping Based Recognition System for Two Dimensional Objects**,

*CVWS87*(164-169).
*Recognize Two-Dimensional Objects*. The lines representing the contours of the overlapping objects
are grouped together. The groups are then recognized. Not as
good as the contour matching programs.
BibRef
**8700**

*van Hove, P.[Patrick]*,

**Model-Based Silhouette Recognition**,

*CVWS87*(88-93).
BibRef
**8700**

And:

**Silhouette-Slice Theorems**,

*CVWS87*(295-297).
*Recognize Two-Dimensional Objects*. A tree-search recognition using the contour edges represented as
line segments.
BibRef

*Hashim, R.*,
*Martin, W.N.*,

**Recognizing Shapes via Random Chord Samplings**,

*CVPR86*(637-639).
BibRef
**8600**

Chapter on Registration, Matching and Recognition Using Points, Lines, Regions, Areas, Surfaces continues in

Partial Contour Matching, Piecewise Segments .

Last update:Dec 7, 2019 at 17:16:29