基于改进的SOFM神经网络的矢量量化方法

基于Kohonen自组织特征映射(SOFM)神经网络的矢量量化图像压缩编码是一种非常高效的方法,但其码字利用不均匀,某些神经元永远无法获胜而产生"死神经元"的问题仍然十分明显。在追求为使各个神经元能以较为均衡的几率获胜,尽量避免"死神经元"过程中,Kohonen SOFM-C很具代表性,它既能保持拓扑不变性映射又能最有效地避免"死神经元",是一种带"良心"的竞争学习方法。本文利用Kohonen SOFM-C码字利用更为均衡的优点,并针对SOFM在胜出神经元的邻域内神经元修改权值方法的不足,提出基于SOFM-C的辅助神经元自组织映射算法,此方法具有开放性,可随时添加入新的有效算法模块以达到更好的效果。并把该矢量量化算法应用于小波变换域,以获得更好的码书。仿真结果表明,该方法优于已有的SOFM方法。

Vector Quantization Based on the Improved Self-Organizing Feature Mapping Neural Networks

The image impression method of vector quantization based on the improved self-organizing feature mapping neural networks is a very efficient way.However,code word cannot be uniformly used.Some neurons never win and the problem of "dead neurons" is still very evident.Kohonen SOFM-C is a conscientious competitive learning method.It can maintain the topological invariant map and avoid "dead neurons" most effectively.In this paper,based on the auxiliary SOFM-C neurons,a self-organizing mapping algorithm is proposed.The method is open and you can always add modules into the new effective method to achieve better results.The vector quantization algorithm has been applied to the wavelet transform domain to obtain a better codebook.The simulation results show that the method is better than the existing SOFM method.