Optimal fuzzy competitive learning self-organizing feature map

In this paper, we introduce a novel optimal fuzzy competitive learning self-organizing feature map (FCL-SOFM.). Different from traditional SOFM, the neurons in FCL-SOFM are updated in the whole neuron lattice based on the fuzzy competitive membership function. So, how to choose membership function is very important in FCL-SOFM. A novel optimal membership function selection scheme is proposed in this paper, in which the fuzzy exponential factor is chosen based upon the normalized Gibbs distribution of network energy in each iterative stage. This optimal FCL-SOFM network can finally converge to the base state of system energy, and achieve global minimum. We apply this optimal FCL-SOFM neural network in Vector Quantization to construct optimal codebook. The experimental result shows that this optimal FCL_SOFM Vector Quantization has a better compressing performance than JPEG.