FLAMINGO aims at detecting and counting flamingos (or any other bird species or object) on any kind of images. We use a stochastic approach based on a marked point process of ellipses. Each ellipse represents one flamingo. The optimization of the model is performed by a new fast birth-and-death dynamics. This leads to the total number of flamingos. The reader can refer to Technical Reports for more details: - X. Descombes, R. Minlos, E. Zhizhina. "Object extraction using a stochastic birth-and-death dynamics in continuum". Research Report, INRIA, 6135, March 2007. - S. Descamps and X. Descombes and A. Béchet and J. Zerubia. Automatic flamingo detection using a multiple birth and death process. Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Las Vegas, USA, March 2008. - Descamps, S.; Béchet, A.; Descombes, X.; Arnaud, A. & Zerubia, J. 2011. An automatic counter for aerial images of aggregations of large birds. Bird Study, 58:302-308
The program was developed in C++ and uses an additional C++ library, the CImg library developed by D.Tschumperle under a CeCILL licence. This library can be downloaded on http://cimg.sourceforge.net/download.shtml.
In the window version parameters input is straightforward through the GUI. 1. Open an input image. Usable formats are png and jpeg.
2. adjust the min and max size (in pixel) of the small and large axis of the ellipse.
3. Possibly change optional parameters: The unique type of contrast in the image: 0 for dark objects on bright background or 1 for bright objects on dark background. The birth rate that is initially fixed to 1. But, if you observe missing detection, you should increase the birth rate to, for example, 2. The last parameter is the number of iterations of the birth-and-death dynamics. You should leave it to 2000.
4. Run. A window open and let you draw the area of the image you want to count (clic left to start drawing). Once you close the window, the program starts counting.
For the linux version you have to modify the parameter file depending on the image. 1. The two first parameters refer to the input image to be analyzed and the output image after the detection where we can see ellipses placed on flamingos.
2. The following three parameters refer to the size of the flamingos searched on the image. The first represents the minimal semi-axe of the ellipses. The second is the pixel freedom of the great semi-axe (from the minimal semi-axe) and the third is the freedom of the small semi-axe (from the minimal semi-axe).
3. Possibly change optional parameters
- the unique type of contrast in the image: 0 for dark objects on bright background or 1 for bright objects on dark background. - the birth rate that is initially fixed to 1. But, if you observe missing detection, you should increase the birth rate to, for example, 2. - The last parameter is the number of iterations of the birth-and-death dynamics. You should leave it to 2000.
To allow us evaluating the interest of wildlife scientists for the product, please enter the following informations:
In each archive two test images are provided by the Tour du Valat: Tuz04.png and Tuz04Result.png which is the result of the automatic detection.
|