thirakawa / SegmentedShapeFeatures

Segmented Shape Features (SSF)

This repository implements a texture feature which is called Segmented Shape Features (SSF) proposed by Häfner et al. [1].

This implementation has a small difference as follows. Häfner et al. have used and simplified Fast Level Lines Transform (FLLT) [2] to construct upper and lower level line sets from an image. While we construct the level line sets by Tree of Shapes (ToS) using an algorithm proposed by Géraud et al. [3].

Execution environment

• OS: Mac OS X (10.9 or 10.11)
• Language: Python 2.7.* (Anaconda 2.4.*)
• Modules: Numpy, Matplotlib, OpenCV, NetworkX, Cython

Usage

python setup.py build_ext --inplace
python SegmentedShapeFeatures.py [image file]

Sample images

Some of the Brodatz dataset images are stored in images directory.
Brodatz dataset

References

1. M. Häfner, A. Uhl, and G. Wimmer, "A Novel Shape Feature Descriptor for the Classification of Polyps in HD Colonoscopy," Lecture Notes in Computer Science (LNCS), pp.205-213, 2014.
2. P. Monasse and F. Guichard, "Fast computation of a contrast-invariant image representation," iin Image Processing, IEEE Transactions on, vol.9, no.5, pp.860-872, 2000.
3. T. Géraud, E. Carlinet, S. Crozet, and L. Najman, "A quasi-linear algorithm to compute the tree of shapes of nD images," International Symposium on Mathematical Morphology (ISMM), pp.98-110, 2013.