Reference image background reference used for change detection segmentation

1
Reference image updating. Overview

• Reference image background reference used for
  change detection segmentation
• Problem
• Not always available
• Subject to local and global variations
  (illumination and shadow changes, )
• Goal of the method
• Learning then updating the reference image
• How it works
• Local variations Pixels of the current image are
  classified stable or not
• if they belong to a mask or a moving zones, these
  pixels are said  not stable . They dont change
  the reference image content at all.
• If they dont, theses pixels are said stable .
  They are taken into account (pixelwise) to
  smoothly change the reference image content.
• Global variations when 2 successive images are
  too much different (fast changes) or when the
  current image becomes too much different compared
  to the reference image (slow change)
• The global difference is applied to the whole
  reference image to correct it.

2
Reference image updating. Block diagram
3
Reference image updating. Results
The reference image is learned and updated,
taking into account wheather changes
After a short learning phase, masks of people are
well extracted