模糊增强学习码书设计算法及其在图像编码中的应用

本文提出一种用于图像压缩编码的模糊增强学习码书设计算法。该算法是在模糊竞争学习矢量量化的基础上引入增强学习,并用输入训练模式的监督信号与类别模式之间的隶属度控制增强信号。实验结果表明,该算法对初始码本依赖性小,与模糊竞争学习矢量量化和微分竞争学习算法相比,收敛速度更快,性能更好。

Design of Codebook with Fuzzy Reinforcement Learning Vector Quantization

A fuzzy reinforcement learning vector quantization (FRLVQ) algorithm was proposed for the optimal codebook design in image compression. Basically, it is a competitive soft learning vector quantization embedded with reinforcement learning, and the reinforcement signal is so constructed as to depend on the membership function of the supervised signal of the corresponding training vector. Experimental results show that, the new algorithm is robust with different initial codebook selection, and converges faster with better performance contrast to the differential competitive learning (DCL) algorithm and fuzzy competitive learning vector quantization (FCLVQ) algorithm.