pvinet

Computes the discrepancy measure between two region maps based on the number of mis-segmented pixels.

Synopsis

pvinet [-m mask] [rg1_in|-] [rg2_in|-]

Description

pvinet computes a discrepancy measure that accounts for the disparity between a segmented image (the region map rg1_in) and a reference image (the region map rg2_in) as defined by L. Vinet^*.

The method consists in determining each pair of regions that have a maximum overlapping ratio between the two region maps and to characterize the discrepancy by the number of pixels that do not participate to the overlapping.

The result measure is in [0..1]. The smaller is the discrepancy measure, the better is the segmentation.

Caution: Regions with label=0 are not considered for computing.

Inputs

rg1_in: a region map (segmented image).
rg2_in: a region map (reference image/ ground truth).

Result

Returns a positive real value in [0,1]. 0 means that all the regions are identical, and 1 that all the regions are totally different.
(Use pstatus to get this value).

Examples

Discrepancy measure between two region maps, one translated from the other:

   pbinarization 80 1e30 tangram.pan i1.pan
   plabeling 8 i1.pan i2.pan
   ptranslation 0 10 i2.pan i3.pan
   pvinet i2.pan i3.pan
   pstatus

C++ prototype

Errc PVinet( const Reg2d &rg1_in, const Reg2d &rg2_in );

Version française

Calcul de la mesure de dissimilarité entre deux segmentations basée sur le nombre de pixels mal segmentés.

Reference

^*JP. Cocquerez, S. Philipp, "Analyse d'images: filtrage et segmentation", Masson, 1995.

Author: Régis Clouard