企业网络与系统集成
课程专题报告选编
清华大学自动化系
《企业网络与系统集成》课程小组
二○○二年六月
《企业网络与系统集成》是一门以讲座和研讨为主的研究生课程,主题是计算机网络的发展及其与控制等学科的交叉,渗透与集成.课程要求同学在学习过程中撰写专题报告和课题立项报告等,希望有助于同学扩展知识和创新思考能力的提高.
本文集是2002年课程中所有专题报告的摘要合集以及课程推荐的两篇文章.
谨向大家表示谢意!
编辑小组
2002年6月
课程报告摘要目录
(排名不分先后)
课程推荐论文
A Novel Control-Theoretic Adaptive Streaming Approach for Stored Video
(Dejian Ye, Zuo Zhang and Qiufeng Wu)
Wavelet-based Smoothing and Multiplexing of VBR Video Traffic
(Dejian Ye, Zixiang Xiong, Huai-Rong Shao, Qiufeng Wu, and Wenwu Zhu)
Image compression using query-based heuristic fast self-organizing feature map
盖家鼎
016112 自研31班
Abstract:
This paper presents a fast heuristic query-based version of kohonen's self-organizing feature map (KSOFM) as learning vector quantization to perform quick compression of still images. The original KSOFM has been found very useful in image compression as learning vector quantization, but taking too much time in learning vector quantization has made it unpractical. A heuristic query-based algorithm is introduced here to the original KSOFM for accelerating the finding of the best matching neuron (cluster center) with smart choice, arrangement of the initial states of the neurons and with their desire states. A new image compression scheme is presented based on this heuristic query-based KSOFM, not only for its excellent compression quality feature, but also for its high compression speed property. Comparison of compression quality with JPEG shows proposed scheme provides improved performance of up to 2.03dB peak SNR over JPEG. Comparison of compression speed with similar scheme using original KSOFM3 shows proposed one provides up to 40% improved speed performance.
Keywords: Neural networks, self-organizing, heuristic query-based algorithm, image compression
中文摘要:
本论文提出一种改进的快速的自组织神经网络以替代最初由kohonen提出的kohonen自组织神经网络.这种改进是基于启发式,询问式算法的.基于这种改进, 本文提出的改进自组织神经网络可以快速的进行矢量量化,并可以更实际地应用到快速的图像压缩中去.

下一页