Davis, L.S.,
Hierarchical Relaxation for Shape Analysis,
PRIP78(275-279).
Discrete relaxation applied to a multilevel grammar specification of
shapes. Apply the relaxation at all levels and eliminate
assignments at upper or lower levels when a subpart or superpart is
removed.
See also Hierarchical Relaxation for Waveform Parsing.
BibRef
7800
Kitchen, L., and
Rosenfeld, A.,
Scene Analysis Using Region-Based Constraint Filtering,
PR(17), No. 2, 1984, pp. 189-203.
WWW Version.
BibRef
8400
Earlier:
DARPA82(230-242).
BibRef
And:
UMD-CS TR-1150, DAAG-53-76C-0138, February 1982.
Discrete relaxation (which is important for parallel
representations (implementations)) applied to a graph matching
problem. Desire to have a complete graph, but this is not
practical due to increased cost (especially for hardware
implementation), therefore use sparse graph - proximity and very
large regions are used. Give initial interpretations, filter on
unary constraints (intrinsic properties). Generate all pairs on the
graph arcs and filter these (use these to filter the node
labels).
BibRef
Jacobus, C.J.,
Chien, R., and
Selander, J.,
Motion Detection and Analysis of Matching Graphs of Intermediate
Level Primitives,
PAMI(2), No. 6, November 1980, pp. 495-510.
Similar to Harlow, Baker, Underwood, Dudani(
See also Aircraft Identification by Moment Invariants. ),
et al. Using features
tends to reduce possibilities of matching elements (unique features
occur), and generally they can be consistently located. This
matching is done at the 2-D level, 3-D is derived later, but the
features can be used in the 3-D descriptions process. Lines,
vertices, and regions are encoded in a graph structure which is
used in the matching. Use highly descriptive unique points to seed
the match and have a set of parallel processes at each seed grow
out and compete or merge with others - apply standard relaxation
(e.g., Waltz:
See also Understanding Line Drawings of Scenes with Shadows. )
filtering to propagate out. The matching method
depends very much on the descriptive method for its power.
BibRef
8011
Blake, A.,
Relaxation Labeling: The Principle of 'Least Disturbance',
PRL(1), 1983, pp. 385-391.
BibRef
8300
Blake, A.,
The Least Disturbance Principle amd Weak Constraints,
PRL(1), 1983, pp. 393-399.
BibRef
8300
Radig, B.M.[Bernd M.],
Image Sequence Analysis Using Relational Structures,
PR(17), No. 1, 1984, pp. 161-167.
WWW Version.
BibRef
8400
Earlier:
Hierarchical Symbolic Description and Matching of Time Varying Images,
ICPR82(1007-1010).
Generate various descriptions using maximal cliques and match these
cliques at the higher levels (cliques to cliques). Use relational
structures for matching. Find cliques and use to find similar
cliques in other image. Clique: completely connected subset.
Group into objects (almost always cliques) and detect similar
cliques in the image. A long formal description of relational
structures, but little on actual examples and results.
BibRef
Radig, B.M.,
Kraasch, R.,
Zach, W.,
Matching Symbolic Descriptions for 3D Reconstruction of
Simple Moving Objects,
ICPR80(1081-1084).
BibRef
8000
Gu, J.,
Wang, W., and
Henderson, T.C.,
A Parallel Architecture for Discrete Relaxation Algorithm,
PAMI(9), No. 6, November 1987, pp. 816-831.
BibRef
8711
Ankenbrandt, C.A.,
Buckles, B.P.,
Petry, F.E.,
Scene Recognition Using Genetic Algorithms with Semantic Nets,
PRL(11), 1990, pp. 285-293.
BibRef
9000
Boyce, J.F.,
Feng, J.,
Haddow, E.R.,
Relaxation Labelling and the Entropy of Neighborhood Information,
PRL(6), 1987, pp. 225-234.
BibRef
8700
Koo, J.Y.,
Park, K.H.,
Kim, M.,
Improving the Labeling Accuracy by a
New Probabilistic Relaxation Labeling,
PRL(3), 1985, pp. 399-402.
BibRef
8500
Kirousis, L.M.[Lefteris M.],
Fast parallel constraint satisfaction,
AI(64), No. 1, October 1993, pp. 147-160.
WWW Version. Motivated by line labeling systems.
BibRef
9310
Cucchiara, R.,
Lamma, E.,
Mello, P.,
Milano, M.,
Piccardi, M.,
3D object recognition by VC-graphs and interactive constraint
satisfaction,
CIAP99(508-513).
IEEE DOI Link
9909
BibRef
Chapter on Matching and Recognition Using Volumes, High Level Vision Techniques, Invariants continues in
Discrete Relaxation Theoretical Issues .