pvarianceaggregation

Performs pixel aggregation based on variance criterion.

Synopsis

pvarianceaggregation connexity threshold [-m mask] [rg_in|-] [im_in|-] [rg_out|-]

Description

pvarianceaggregation builds a new region map from aggregation of connected pixels to regions of the input region map rg_in. A pixel p is aggregated to a region R if:

p is connected to the region R according to the specified connexity;
|variance(R)-variance(R+im_in[p])| <= threshold

The variance degree is calculated from:

    variance(R)= SUM((im_in[i] - mean(R))^2, i in R) / N

where im_in[i] are pixels of the region R, and N is the number of pixels of the region R.

The variance of the region is not updated with the new pixel to avoid moving away too much from the initial situation. One prefer iterative executions of the operator to update the inner mean. For example, operator can be iterated until pstatus returns 0.

The output region map rg_out has the same number of labels than the input region map.

Parameters

connexity specifies the neighbor relation between pixels (4 or 8 for 2D; 6 or 26 for 3D).
threshold specifies the maximum variance value to decide to aggregate a pixel to the region. Values are from the gray scale of the input image.

Inputs

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

Outputs

rg_out: a region map.

Result

Returns the number of aggregation or FAILURE.

Examples

Aggregates pixels to tangram pieces:

   pbinarization 96 1e30 tangram.pan a.pan
   plabeling 8 a.pan b.pan
   pvarianceaggregation 8 1 b.pan tangram.pan out.pan

C++ prototype

Errc PVarianceAggregation( const Reg2d &rg_in, const Img2duc &im_in, Reg2d &rg_out, int connexity, Uchar threshold );

Version française

Croissance des régions d'une carte selon la variance intérieure.

Author: Régis Clouard