摘要
本文针对图像矢量量化存在的分块效应问题,通过对Kohonen自组织模型的研究,修改了Kohonen的自组织特征映射(SOFM)算法,设计了两个DCT(离散余弦变换)域的特征值,用于图像数据块的分类。在此基础上,进一步探讨了改进的自组织特征映射(MSOFM)算法在图像自适应矢量量化中的应用。计算机模拟实验表明,MSOFM算法有效地减少了分块效应,与SOFM算法相比具有更好的性能。
Based on the discussion of the principle of Kohonen's self-organizing feature maps(SOFM) a modified SOFM (MSOFM) algorithm is proposed to reduce blocking effect of vector quantization of images in this paper. Two eigenvalues are designed in DCT (Discret(?) Cosine Transform) domain to classify image blocks, then we discuss the application of MSOFM algorithm in adaptive vector quantization. The results of computer simulation show that the MSOFM training algorithm significantly reduces blocking effect and have a better Performance than the SOFM algorithm.
出处
《通信学报》
EI
CSCD
北大核心
1992年第5期16-21,共6页
Journal on Communications
基金
国家自然科学基金
关键词
KOHONEN网络
矢量量化
神经网络
Kohonen network, Vector quantization, Neural network, Image compression, Adaption.
