论文部分内容阅读
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。
In this paper, a multi-resolution image coding scheme based on wavelet transform and neural network is proposed. The wavelet multiresolution representation of the image is used to eliminate the correlation between the image spatial domain and the frequency domain. Since the complex relationship between adjacent lines of wavelet images is difficult to describe by linear expressions, MLNN is used to determine this unknown relationship. Experiments show that the performance of neural network nonlinear predictor is better than that of linear predictor. The non-linear predictive difference image is encoded by a self-organizing feature map (SOFM) codebook in vector quantization (VQ). The coded image has good subjective quality, high compression ratio, simple algorithm and easy implementation in hardware.