Labeling Demo Copyright (c) Sergey Zavalishin 2010-2015

This is a demo application for “Block equivalence algorithm for labeling 
2D and 3D images on GPU” paper, which demonstrates GPU-based block equivalence 
labeling algortihm for 2D and 3D cases. It also contains several other 
algortihms, which were implemented during development of BE algorithm.

The demo implements the following algorithms:

[1] S. Zavalishin, I. Safonov, Y. Bekhtin, I. Kurilin, “Block equivalence 
	algorithm for labeling 2D and 3D images on GPU”, Electronic Imaging 2016, 
	Visual Information Processing and Communication Conference
[2] Grana, Costantino, Daniele Borghesani, and Rita Cucchiara. "Optimized 
	block-based connected components labeling with decision trees." Image 
	Processing, IEEE Transactions on 19.6 (2010): 1596-1609.
[3] He, Lifeng, Yuyan Chao, and Kenji Suzuki. "A run-based two-scan labeling 
	algorithm." Image Processing, IEEE Transactions on 17.5 (2008): 749-756.
[4] Y. Bekhtin, V. Gurov, S. Zavalishin, “A Run Equivalence Algorithm for 
	Parallel Connected Component Labeling on CPU”, Embedded Computing (MECO), 
	2015 4th Mediterranean Conference, pp.276-279
[5] Kalentev, Oleksandr, et al. "Connected component labeling on a 2D grid 
	using CUDA." Journal of Parallel and Distributed Computing 71.4 (2011): 
	615-620.
[6] Wu, Kesheng, Ekow Otoo, and Kenji Suzuki. "Two strategies to speed up 
	connected component labeling algorithms." Lawrence Berkeley National 
	Laboratory (2008).

Notes:

[2] is based on the code from http://phaisarn.com/labeling
[3] is a relatively slow version, which was implemented for the reference only
[6] is taken from OpenCV 3.0.0 implementation, thus this algorithm may be
    changed in further versions of OpenCV

Compilation:

The demo requires the following third party components:

	* OpenCV 3.0.0
	* boost 1.55.0
	* OpenCL 1.1

Originally it was designed for OpenCL headers from CUDA 7.0 , but there's
no specific restrictions on using any other compatible headers from other
OpenCV vendors.

Usage:

Detailed help is embedded into the demo. Use '-h' flag to get it.
The basic scenario includes labeling for the single image, which may be
performed using the following flag combination:

	labeling -a bleq -g -i "image.png" -o "out_path"

It means that image.png will be labeled using GPU version of Block 
Equivalence algorithm [1] and the result stored at "out_path". Note, 
that storing labeled image is a time consuming process, thus try to 
avoid using it.

Another scenario includes batch processing:

	labeling -a gr-block -i "in_path"

Using these flags you can label all images from "in_path" with Grana
alrotihm [2].

Finally, you may label 3D images stored as a set of 2D slices:

	labeling -a lbeq -3 -g "in_image"

These flags mean that input image from "in_image" directory will be
labeled using Label Equivalence algortihm [5] on GPU.