psuperpixelsegmentation

Synopsis

Description

This operator computes an over-segmentation of the input image im_in. The initial seeds are taken from the region map rg_in. Seeds are regions of size 1. The algorithm is based on region growing clustering using the Eikonal equation. The resulted image im_out stores the geodesic distance map.

A spatial constraint term can be used to enforce compactness of the superpixels.

Note: for color image, results are better in the lab color space.

Parameters

• compactness is a factor that impacts the shape of the superpixels. It is a real value between 0 and 1. The default value is 0.
• perturbation is a boolean that specifies whether it is possible to move the seed (perturbation=1) or not (perturbation=0).

Inputs

• reg_in: a region map with the initial seeds.
• im_in: an image of intensity.

Outputs

• reg_out: a region map.
• im_out: the geodesic distance map.

Result

Returns SUCCESS or FAILURE.

Examples

• Place seed every 10 pixels on the grid then build the superpixels around the seeds:

  ./pseedplacement 10 10 0 examples/tangram.pan rgin.pan
  ./psuperpixelsegmentation 0 1 rgin.pan examples/tangram.pan rgout.pan geodesic-map.pan
  pboundary 8 rgout.pan cont.pan
  psuperimposition 3 cont.pan examples/tangram.pan result.pan
  

See also

C++ prototype