R. Xiao, C.-H. Chang, and T. Srikanthan (Singapore)
Self-Organizing Feature Maps, ColorQuantizaiton, Artificial Neural Network.
Kohonen’s self organizing feature maps (SOFMs)
have been used for color quantization in various image
compression algorithms. Intuitively, the cause of slow
convergence and poor generalization ability exhibited
by some conventional SOFMs can be attributed to the
ignorance of the localization of the adaptive processes.
In this paper, a new learning algorithm is proposed
where the rate of adaptation is based on the localized
winning frequency of each individual neuron.
Simulation results show that the proposed frequency
adaptive learning algorithm can achieve a better
reconstructed image quality and speed up the
convergence by eight times for a 256-neuron network.