SGONG-Color Reduction

Color Reduction using a self growing and self organized neural network

This color reduction software is based on a new neural network classifier (SGONG). The SGONG network classifier is suitable for color quantization applications, since each image pixel is considered as multidimensional vector, depending on the color space that is used and the extra features which are derived from the neighborhood domain of each pixel. The main advantage of SGONG network is that it controls the number of created neurons and their topology in an automatic way. This advanced feature makes feasible the automatic extraction of the principal colors of an image. The convergence speed of SGONG classifier is comparable to the convergence speed of the GNG classifier, while the stability of SGONG classifier is comparable to the stability of Kohonen SOFM classifier. In order to speed up the entire algorithm, a fractal sub-sampling procedure based on the Hilbert's space filling curve is applied to initial image, taking samples only from the fractal peaks and their neighboring pixels.