【计算机科学】【2003.08】基于级联神经网络的图像压缩

本文为美国新奥尔良大学（作者：Chigozie Obiegbu）的硕士论文，共108页。

图像正逐渐成为现代通信的一个重要组成部分，从而需要进行高效且有效的压缩。为此目的开发的许多技术包括变换编码、矢量量化和神经网络。本文采用一种新的神经网络方法来实现图像压缩，这项工作将两层神经网络的使用扩展到级联网络与隐藏层中一个节点的组合。训练阶段的灰度再分配是以随机方式实现的，以使均方误差最小化，适用于广泛的多种图像。从权值的总个数和总体收敛性角度分析了该方法的计算复杂性。图像质量的测量是基于客观方式的峰值信噪比和主观的知觉感受。评价了不同图像内容和压缩比对图像质量的影响。结果表明，级联神经网络在高压缩比下的性能优于固定结构训练模型。该方法在MATLAB中予以实现，给出了压缩比和压缩图像的计算时间等结果。

Images are forming an increasingly largepart of modern communications, bringing the need for efficient and effectivecompression. Many techniques developed for this purpose include transformcoding, vector quantization and neural networks. In this thesis, a new neuralnetwork method is used to achieve image compression. This work extends the useof 2-layer neural networks to a combination of cascaded networks with one nodein the hidden layer. A redistribution of the gray levels in the training phaseis implemented in a random fashion to make the minimization of the mean squareerror applicable to a broad range of images. The computational complexity ofthis approach is analyzed in terms of overall number of weights and overallconvergence. Image quality is measured objectively, using peak signal-to-noiseratio and subjectively, using perception. The effects of different imagecontents and compression ratios are assessed. Results show the performancesuperiority of cascaded neural networks compared to that of fixedarchitecturetraining paradigms especially at high compression ratios. The proposed newmethod is implemented in MATLAB. The results obtained, such as compressionratio and computing time of the compressed images, are presented.

1 引言
2 图像压缩技术
3 人工神经网络技术回顾
4 基于神经网络的图像压缩
5 结果
6 讨论与结论