Inverse problems, suchasthereconstruction problems thatarise inearly vision, tendtobemathematically ill-posed. Through regularization, theymaybereformulated aswell-posed variational principles whosesolutions arecomputable. Standard regularization theory employs quadratic stabilizing functionals thatimpose global smoothness constraints onpossible solutions. Discontinuities present serious difficulties tostandard regularization, however, since their re- construction requires a precise spatial control overthesmoothing properties ofstabilizers. Thispaperproposes ageneral class ofcon- trolled-continuity stabilizers whichprovide thenecessary control over smoothness. Thesenonquadratic stabilizing functionals comprise mul- tiple generalized spline kernels combined with(noncontinuous) conti- nuity control functions. Inthecontext ofcomputational vision, they maybethought ofascontrolled-continuity constraints. Thesegeneric constraints areapplicable tovisual reconstruction problems thatin- volvebothcontinuous regions anddiscontinuities, forwhichglobal smoothness constraints fail.
[1]
A. Booth.
Numerical Methods
,
1957,
Nature.
[2]
V. K. Ivanov,et al.
On numerical differentiation
,
1966
.
[3]
C. Reinsch.
Smoothing by spline functions
,
1967
.
[4]
R. N. Desmarais,et al.
Interpolation using surface splines.
,
1972
.
[5]
David Marr,et al.
Representing Visual Information
,
1977
.
[6]
G. Wahba,et al.
Some New Mathematical Methods for Variational Objective Analysis Using Splines and Cross Validation
,
1980
.
[7]
Andrew P. Witkin,et al.
Scale-Space Filtering
,
1983,
IJCAI.
[8]
J J Hopfield,et al.
Neurons with graded response have collective computational properties like those of two-state neurons.
,
1984,
Proceedings of the National Academy of Sciences of the United States of America.
[9]
J. Marroquín.
Surface Reconstruction Preserving Discontinuities
,
1984
.
[10]
Alan L. Yuille,et al.
A regularized solution to edge detection
,
1985,
J. Complex..