puniformitymerging

Performs priority region merging based on uniformity criterion.

Synopsis

Description

puniformitymerging merges connected regions of the input image rg_in if the difference between the uniformity of the region is lower than the specified threshold.

Two regions are connected if there exists a link between the related nodes in the input graph gr_in.

The principle of the algorithm is as follows:

For each region of the input region map rg_in,
If the difference between the criterion value of the two connected regions ≤ threshold
then merge them into one region.

The algorithm uses the priority merging that consists in merging regions with the lower difference.

The output region map reg_out defines the new regions and the output graph graph_out defines the new relationship between regions.

The uniformity criterion is calculated from:

   uniformity(R)= 1 - ( variance(R) / (mean(R)^2) )

Parameters

number specifies the number of allowed merging. If number = -1 then all possible merging are done.
threshold specifies the maximum difference allowed between two regions to decide to merge them. Values are from [0..1], where 0 corresponds to non uniform regions, and 1 corresponds to highly uniform regions. A typical value is 0.95.

Inputs

rg_in: a region map.
gr_in: a graph.
im_in: a grayscale image.

Outputs

rg_out: a region map.
gr_out: a graph.

Result

Returns the number of merging.

Examples

Merges regions yielded by a quadtree splitting process:

   puniformityquadtree 0.9 tangram.pan a.pan
   prg2gr a.pan b.pan
   puniformitymerging -1 0.99 a.pan b.pan tangram.pan c.pan d.pan

C++ prototype

Errc PUniformityMerging( const Reg2d &rg_in, const Graph2d &gr_in, Img2duc &im_in, Reg2d &rg_out, Graph2d &gr_out, long number, float threshold );

Version française

Fusion prioritaire de régions selon le critère d'uniformité.

Author: Laurent Quesnel