Overview

The scikit-gtimage (standing for ‘’graph tools for image analysis’‘), skgtimage for package aims at providing graph-based tools for image interpretation. It implements a method, based on an inexact graph matching technique, for retrieving and identifying a set image regions, from an initial oversegmentation, using a priori declared qualitative inclusion and photometric relationships between the set of image regions. This package is distributed under the 3-Clause BSD license.

Downloads

  • Documentation
  • Code
  • Database

    • Principle of the approach

    Mountain View Analysis of an input segmented image (labelled regions 0 to 5), using an a priori model (relationships regarding expected regions A, B, C and D). Relationships regard inclusion (e.g. B in A) and photometric relationships (e.g. D brighter than C). Labelled regions 2 and 4 represent artefacts that can be encountered in real images. Based on automatically built relationships, one first searches for subgraph isomorphisms (between model and built graphs) that are common to both inclusion and photometric relationships. According to a selection criterion, only one is considered (represented by red arrows). Remaining unmatched nodes are finally matched (green arrows) for recovering and identifying regions. Both red and green arrows represent the inexact graph matching