关键词: 独立分量分析 负熵最大化 超高斯 矢量量化

Abstract:

On the basis of studying ICA extensively, we apply the FastICA algorithm based on negentropy maximition to number recognition. After the new characters are aquired, an algorithm of improved LVQ is used for recognition. The experiments in the number recognition show that the ICA method has a higher recognition rate than the PCA method, and a lower computing complexity than tradational recognization methods.

Key words: ICA, negentropy maximization, supergaussian, LVQ