引入神经网络中间神经元的快速小波图像压缩 - 中国图象图形学报.doc

中图法分类号：TP751.1 文献标识码：A 文章编号： 论文引用格式： 引入神经网络中间神经元的快速小波图像压缩 张海涛1， 张永霖1 1.辽宁工程技术大学软件学院，葫芦岛 125105 摘 要：目的 针对自组织特征映射(SOFM)算法会出现严重的分块现象和快速小波变换在高压缩比的情况下图像恢复质量差的问题，提出引入神经网络中间神经元的RSOFM-C矢量量化算法。方法 引入了中间神经元 (Relay neurons)的概念，DPCM)中的差值信号编码原理将RSOFM-C算法与快速小波变换结合，RSOFM-C算法对由快速小波变换得到的图像低频信号进行进一步压缩。结果 在仿真实验中，将本文算法与同类压缩方法进行对比，当压缩比为52%时本文方法的峰值信噪比(PSNR)达到了39.28，远远高于其他方法。结果表明，本文的压缩算法消除了分块现象，并且在保证高压缩比的同时获得高质量的重构图像。结论 实验表明，本文提出的引入了中间神经元的快速小波压缩方法，具有高压缩比、高保真、速度快等优点，可以高效的压缩图像。 关键词：图像压缩；中间神经元；快速小波变换；神经网络；自组织特征映射 Fast wavelet image compression introducing neural network of relay neuron Zhang Haitao, Zhang Yonglin 1.School of Software，Liaoning Technical UniversiHuludao，125105 Abstract: Objective There would be a serious block phenomenon in SOFM algorithm，and a poor image restoration quality when fast wavelet transform is in the case of the high compression ratio. Method To solve the above problem，RSOFM-C vector quantization algorithm is proposed，in which neural network relay neurons are introduced. The use of relay neurons solves the problem of uneven code words by introducing the concept of relay neurons，uclidean distance discriminant inequality is given in neural network middle layer，neurons that do not satisfy the distortion measure are excluded，it reduces the repeated calculation and accelerate the learning speed. According to the difference signal coding principle in DPCM，the SOFM-C algorithm and the fast wavelet transform are combined. The low frequency image signal is further compressed by using the RSOFM-C algorithmResult In the simulation experiment，the algorithm in this paper is compared with similar compression method. When the compression ratio is 52%the peak signal to noise ratio of this method (PSNR) reached 39.28，which is much higher than other methods. The compression algorithm in this paper eliminated the blocking phenomenon，and high quality reconstructed image can be obtained while ensuring the high compression ratio. Conclusion Experiment showed that by introducing t