next up previous contents index
Next: drawocclusion Up: Reference Previous: cdisc   Contents   Index


$ \bigcirc$Name

disocclusion Disocclusion using global minimisation of cost by dynamic recursive programing

$ \bigcirc$Command Synopsis

disocclusion [-e energy_type] [-a] input holes output

-e energy_type : Energy of a level line : 0 = only length, 1 = only angle (default), otherwise = angle+length

-a : If used then the orientation of each entering level line is computed more accurately on a ball of radius 4

input : Input occluded Cimage

holes : Input Fimage containing the only occlusions

output : Output disoccluded Cimage

$ \bigcirc$Function Summary

void disocclusion (Input , Output , Holes , energy_type , angle )

Cimage Input , Output ;

Fimage Holes ;

char *angle ;

int *energy_type ;

$ \bigcirc$Description

This modules performs a singular (level-lines based) interpolation of missing parts in the Input image. There is no detection of missing parts, which are assumed to be completely specified by the Holes image, each of them being uniquely associated with a positive value (the drawocclusion module can be used to create such a mask image of the holes). Missing parts must be simply connected domains (i.e, without hole). In case they are not, the algorithm will automatically change them into simply connected sets by simply adding the holes.

The interpolation is carried out in the following way : for each missing part, the algorithm finds an optimal set of interpolating level lines {Li, ti $ \in$ Itt $ \in$ IR} according to the energy

$\displaystyle \int_{{-\infty}}^{{+\infty}}$($\displaystyle \sum_{{i\in I_t}}^{}$$\displaystyle \int_{{L_{i,t}}}^{}$($\displaystyle \alpha$ + $\displaystyle \beta$|$\displaystyle \kappa$|)d$\displaystyle \cal {H}$1)dt

where $ \alpha$,$ \beta$ $ \geq$ 0, $ \kappa$ denotes the curvature and $ \cal {H}$1 is the one-dimensional Hausdorff measure. The term $ \int_{{L_{i,t}}}^{}$|$ \kappa$| d$ \cal {H}$1 actually represents the angle total variation along Li, t, taking into account the angles at both endpoints.

The option -a allows to compute more accurately the directions of the outer level lines at the boundary of each missing part.

The values of $ \alpha$ and $ \beta$ depends on the parameter energy_type. More precisely

energy$\displaystyle \_type$ = $\displaystyle \left\{\vphantom{\begin{array}{lll}
1\mbox{ (default)}\\
\mbox{otherwise}\end{array}}\right.$$\displaystyle \begin{array}{lll}
1\mbox{ (default)}\\
\mbox{otherwise}\end{array}$ $\displaystyle \Rightarrow$ $\displaystyle \left\{\vphantom{\begin{array}{lll}
\alpha=0,\;\beta=1\mbox{ (default)}\\
\alpha=0.2,\;\beta=1\end{array}}\right.$$\displaystyle \begin{array}{lll}
\alpha=0,\;\beta=1\mbox{ (default)}\\

The minimization is performed by a dynamic programming approach and a global minimal solution is computed. See [Mas02] for a detailed description of the algorithm.

$ \bigcirc$See Also


$ \bigcirc$Version 2.0

Last Modification date : Thu Apr 15 04:45:34 2004

$ \bigcirc$Author

Simon Masnou

next up previous contents index
Next: drawocclusion Up: Reference Previous: cdisc   Contents   Index
mw 2004-05-05