摘要
本文提出了一种基于小波变换与神经网络的多分辨率图像混合编码方案，利用小波分解对图像的多分辨率表示来消除图像空间域和频率域的相关性。由于小波图像相邻行之间的复杂关系难以用线性表达式来描述，使用多层神经网络（ＭＬＮＮ）来确定这种未知关系。实验证明，神经网络非线性预测器性能优于线性预测器。对非线性预测后的差值图像用自组织特征映射（ＳＯＦＭ）码书进行矢量量化（ＶＱ）编码，编码图像主观质量好，压缩比高，算法简单，易于硬件实现。
A multiresolution image hybrid coding scheme based on wavelet transform and neural networks is presented in this paper.The statistical correlation of the images in both space and frequency domains is eliminated by the multiresolution representation of the images provided by wavelet decomposition.Since linear expressions are not suited for the complex relationship between the adjacent scanning lines of the wavelet images,this paper uses multi-layer neural network(MLNN)to determine the unknown relationship.It has been demonstrated that the performance of nonlinear predictor using neural networks is better than linear predictor.The displaced residual images after nonlinear prediction are VQ coded with self-organizing feature maps(SOFM) codebook.The subjective quality of the coded images is very satisfactory with high compression ratio.The coding scheme is simple and very easy to implement with hardware.
出处
《通信学报》
EI
CSCD
北大核心
1995年第2期71-78,共8页
Journal on Communications
基金
国家自然科学基金
中科院自动化所国家模式识别实验室资助
关键词
混合编码
小波变换
神经网络
图像编码
multiresolution image
hybrid coding
wavelet transform
neural networks
