In order to save a temporary snapshot of an image you should press the “snapshot” button, give the corresponding name and press “Enter”.
These temporary images can be used as extra features for an image, or you can take comparative results between them.
Three color reduction algorithms can be applied. These techniques are based on three neural network classifiers, respectively.
The following neural network classifiers are used for color reduction
Kohonen Self Organized Feature Map (Kohonen SOFM) [1], [2], [5], [6]
Growing Neural Gas (GNG) [3], [4], [5], [6]
Self Growing and Organized Neural Gas (SGONG)
References
[1] T. Kohonen, "The self-organizing map," Proceedings of IEEE vol. 78, no. 9, pp.1464-1480, 1990
[2] T. Kohonen, Self-Organizing Maps. 2nd Edition, Springer Verlag, Berlin, 1997
[3] B. Fritzke, "Some competitive learning methods", Draft document, http : //www. neuroinformatik.ruh- uni-bochum.de/ini/VDM/ research/gsn/DemoGNG, 1998.
[4] B. Fritzke, "A growing neural gas network learns topologies," in Advances in Neural Information Processing Systems 7, G. Tesauro, D. S. Touretzky, and T. K. Leen, Eds. Cambridge, MA: MIT Press, 1995, pp. 625-632.
[5] A. Baraldi and P. Blonda, “A Survey of Fuzzy Clustering Algorithms for Pattern Recognition—Part I”, IEEE Transactions On Systems, Man, And Cybernetics—Part B: Cybernetics, Vol. 29, No. 6, pp. 778-785,December 1999.
[6] A. Baraldi and P. Blonda, “A Survey of Fuzzy Clustering Algorithms for Pattern Recognition—Part II”, IEEE Transactions On Systems, Man, And Cybernetics—Part B: Cybernetics, Vol. 29, No. 6, pp. 786--801, December 1999.
Except the RGB color space, which is not perceptually uniform, more advantageous and perceptually uniform color spaces like CIE-L*a*b* or CIE-L*u*v*, can be used.
The dimensionality of the input space can be increased assigning to each pixel (input vector ) extra features which correspond to the color information of the corresponding pixel of already saved images (snapshots) with the same size.
The common settings in order the final quantized image has exactly 16 colors are:
m1=0, m2=0, “maximum number of colors”=16
The common settings in order that the final quantized image has at most 16 colors are:
m1=0.4, m2=0.3, “maximum number of colors”=16
Load the following color image. This image has 33806 unique colors.
Save the initial image on snapshot's list, giving the name “Original image”
Apply the SGONG, GNG and Kohonen SOFM color reduction algorithms, saving in snapshot's list the resultant images, as depicted in the following figure:
In this example the initial image is quantized to 8 colors using the following settings
The resultant images are:
Using the SGONG network
Using the GNG network
Using the Kohonen SOFM network
Press the “Compare Images” menu item to appear the following form. In order to compare the snapshots “Original_Image” with “SGONG_8_colors”, “GNG_8_colors” and “Kohonen_SOFM_8_colors”, respectively, fill the List1 and List2 as follows.
Press the “Compare” button to create and open the “results.txt” file which contains the comparative results.
