The Self-Organizing Mapping Based on Kernel Function

Distance measurement function of SOM has great impact on learning process, it directly affects the matching degree between the winning neuron and the current input neuron. The Euclidean distance measurement usually taken makes learning process achieve success easily for date set with Euclidean hypersphere distribution. But for other distributions of date sets, the distance measurement may lead to bad results. Therefore, the distance measurement based on kernel function is applied in SOM (KSOM) in this paper. And three image data sets are chosen for investigating KSOM's effect: Lena, Boat, and Peppers. Experiment results show that the modified SOM has great results for date set with non-Euclidean hypersphere distribution.