Interactive segmentation with morphological hierarchies

Watershed
The watershed forms superpixels from the full resolution image. This enables to decrease the complexity of the induced hierarchy at the cost of reduced possibilities (default: Off).
Color gradient
- L2 RGB
- L2 L*a*b*
The function that measures the dissimilarity between two colors (default: L2 L*a*b*).
Flooding by attribute
- Volume
- Area
- Disabled
The original hierarchy can be reweighted following a measure of the importance of the catching bassins. Select a measure to reweight the hierarchy or "Disabled" to work on the original hierarchy (default: Volume).
Adjacency
Adjacency describes the neighbooring relation between pixels (default: 8).

Technical information

This work is an implementation of a marked segmentation approach with Hierarchies of minimum spanning forests.

Hierarchies of minimum spanning forests and algorithms to construct them are described in the following articles:

J. Cousty, L. Najman, B. Perret. Constructive links between some morphological hierarchies on edge-weighted graphs. 11th International Symposium on Mathematical Morphology, ISMM 2013. doi:10.1007/978-3-642-38294-9_8
L. Najman, J. Cousty, B. Perret. Playing with Kruskal: algorithms for morphological trees in edge-weighted graphs. 11th International Symposium on Mathematical Morphology, ISMM 2013. doi:10.1007/978-3-642-38294-9_12
J. Cousty, L. Najman. Incremental algorithm for hierarchical minimum spanning forests and saliency of watershed cuts. 10th International Symposium on Mathematical Morphology, ISMM 2011. doi:10.1007/978-3-642-21569-8_24

The marker based segmentation method is described in:

P. Salembier and L. Garrido. Binary Partition Tree as an Efﬁcient Representation for Image Processing, Segmentation, and Information Retrieval, IEEE Transaction on Image Processing, 2000, 9(4):561-576. doi:10.1109/83.841934.
B. Perret, J. Cousty, O. Tankyevych, H. Talbot and N. Passat. Directed connected operators: asymmetric hierarchies for image filtering and segmentation, Technical Report, 2014.

Technical limitation

Although the algorithms used in this work are efficient, the javascript overhead imposes a strong limitation on the maximum image size. Currently any image larger than 600x600 pixels will be downsampled automatically to fit within these bounds.

About the author

My name is Benjamin Perret. I am a teacher-researcher in computer science at the engineering school ESIEE Paris of the Université de Paris Est. I am also a member of the Laboratoire d'Informatique Gaspard-Monge (LIGM).

More details on my research interests and teachning activity are available in my personnal web page.

For any question, comment, suggestion, or bug repport, feel free to send me a mail .

License

Except where otherwise noted, this website is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. This means that you can freely use and modify any part of it for non commercial uses under the condition that you give appropriate credit (by putting a link to this page), provide a link to the license, and indicate if changes were made.

The code/software hosted in this website (aka the javascript scripts) is licensed under CeCILL (a GPL compatible licence).

The demonstration images come from the Grabcut database.

This work uses the following libraries:

Drag and Drop your image here ! Or

Interactive segmentation: preprocessing in progress...

1. Left click on the object to preserve it.

2. Right click on the background to remove it.

Working

Settings... somewhat advanced!