摘要
随着信息高速公路、数字地球概念的提出以及Internet的广泛使用,信息传输量急剧增长,其中图像信息以其信息量大等一系列优点使其成为人类获取信息的重要来源以及利用信息的重要手段。大量的图像信息需要存储和传输,仅依靠提高信道带宽和计算机的处理速度,不能满足人们对图像信息存储和传输的需要,这就需要结合图像的压缩编码技术来满足人们的要求。当前图像压缩技术在生物医学应用、无线通信、计算机图形图像处理等许多方面有着广泛的应用。
本论文以湖北省科技厅项目“智能运输系统视频信号采集系统识别算法研究”为背景,针对图像信号采集和实时传输中的图像压缩问题,基于图像编码理论、人工神经网络理论和小波变换理论对静态图像的有损压缩技术进行了较深入的分析、比较和研究,提出了静态图像的有损压缩方案。方案首先利用小波多分辨分析性质,对图像进行小波分解,对分解后各子图的小波系数进行了统计分析,针对各子图的小波系数特点,对不同的子图分别采用不同的压缩方法,低频子图采用基于神经网络的自适应预测编码,高频子图采用基于神经网络的矢量量化编码,从而实现对图像数据的压缩处理。
本论文第一章介绍了数字图像压缩处理的国内外当前的概况以及其技术标准和分类。在第二章,介绍了数字图像的矢量量化技术的数学思想和过程,对LBG算法和基于SOFM神经网络的矢量量化进行了阐述、分析。第三章论述了无损和有损预测编码,分析了自适应预测编码中的BP神经网络,论述了神经网络在图像压缩中的算法评估。第四章讨论了数字图像变换域技术在图像压缩中的应用并做了相应的实验,同时对图像变换编码的策略进行了讨论。第五章用小波变换对图像进行分解和表述,讨论了小波函数在图像压缩中的性质和影响,对图像变换后的小波系数进行了分析。建立在前面各章的理论和分析的基础上提出了静态图像的有损压缩方案,并给出了实验结果。实验结果证明了方案的可行性,同时该方案能更好的与图像的局部内容相匹配,去除冗余,获得较好的图像压缩效果。
As the appearance of notion of information highway and digital globe and Internetworks develop widely,the transmission of information data increases rapidly,among these data,image information is the important means and source for people to acquire information and utilize information by its bigger quantity of data. Because many image information data need to be transmitted and preserved, it is difficult to meet the requests for image information data to be transmitted and preserved by depending only on increasing the width of signal band and the processing speed of computers, we need to use image compression techniques to help to satisfy the requests. In these days,image compression is put into practice widely in many fields such as biomedical applications,wireless communications,computer graphics etc.
The paper is under the background of "The research on recognition algorithm of video signal sampling system in intelligent transportation system", which is a project of Hubei province scientific deparment. Pointed to the problem of image compression in image signal sampling and transmitting real time, this paper mainly discusses and analyzes lossy compression of still images deeply based on theories of image coding , artificial neural network , wavelet transform, the paper presents a compression scheme for still images. First the image is decomposed at different scales by using the wavelet transform, then the different quantization and coding schemes for each subimage are carried out according to its statistical properties and distributed properties of the coefficients. The wavelet coefficients in high frequency subimages are compressed and vector quantized based on neural network. The wavelet coefficients in low frequency bands, here adopts a scheme of image coding based on nonlinear predictor by using neural network, finally image is compressed.
In the paper,Chapter 1 gives a comprehensive introduction of digital image compressing including its recent status , technical standards , classification in the world.Chapter 2 introduces briefly the thought and
II
procedure of vector quantization,describes LGB algorithm and vector quantization based on SOFM neural network.Chapter 3 discusses predictable coding in lossy and lossless aspects,analyzes adaptive predictable coding based on BP neural network,introduces the evaluation of algorithm on neural network in image compression.Chapter 4 discusses the applications of mathematical transformation in image compression and does experiments related, analyzes the strategies of image coding in transformed domain.In Chapter 5 images are decomposed and represented by wavelet transform, then discusses the characteristics and effects of wavelet functions in image compression,analyzes the wavelet coefficients after images are decomposed; Based on the theories and analyses in the prior chapters,the paper presents an image compression scheme and gives results.The test results shows that the image compression scheme is practical and helpful to map into the local content of images to get rid off redundancy,so that ,it can require satisfactory results of image compression.
引文
[1] 何振亚.多维数字信号处理.北京:国防工业出版社,1996
[2] 孙即祥.数字图像处理.石家庄:河北教育出版社,1993
[3] 崔屹.数字图像处理技术与应用.北京:电子工业出版社,1997
[4] 张贤达.现代数字信号处理.北京:清华大学出版社,1996
[5] 陈逢时.子波变换理论及其在信号处理中的应用.北京:国防工业出版社,1998
[6] 秦前清,杨宗凯.实用小波分析,西安:西安电子科技大学出版社,1995
[7] 崔锦泰.小波分析导论.西安:西安交通大学出版社,1995
[8] Luttrell S P. Image Compression Using Neural Networks. In:Proc. IGARSS'88,Sept. 1988,pp. 1231-1238,Edinburgh,Scotland
[9] 刘忠贵,邸双亮.小波分析及其应用.西安:西安电子科技大学出版社,1995
[10] 雷寒生,朱小平.基于小波变换和矢量量化的人脸图像压缩.中国图像图形学报,2002,7(1):42—50
[11] Ahalt S C ,Chen P, Melton D E. Competitive Learning Algorithms for Vector Quantization Techniques. Neural Networks, vol. 3,pp.277-290,1990
[12] 孙海威,谈新权.基于小波分形的图像压缩编码.华中科技大学学报,2001,29(2):30—32
[13] 崔东亮,柳斌.分块自适应零树量化编码图像压缩算法研究.华中科技大学学报,2001,29(5):70—80
[14] 马社祥,刘忠贵,尚赵伟.基于小波变换的图像和视频压缩编码.工程数学学报,2001,18(5):18—30
[15] 朱广进,祝建.二维非均匀取样图像编码.上海大学学报,1999,5(4):23—26
[16] 韦长江,郝朋威,石清云.基于整型DCT变换的图像编码研究.中国图像图形学报,2002,7(3):286—292
[17] 虞正华,周功禹.CT图像小波变换系数的统计特征.医学工程学报,1999,2(18)
[18] Cohen A.Doubechies,Feauveau I.Biorthogonal bases of compacted supported wavelets. Commun Pure and Appl Math, 1992,38(2):485-560
[19] AlanC.Bovik, A.Ligtenberg and G.Moschytz. DCT-DOMAIN BLIND MEASUREMENT OF BLOCKING ARTIFACTS IN DCT-CODED IMAGES. Proc. Intll. Conf. on Acoustics, Speech and Signal Processing, 1992(ICASSP92), pp988-991
[20] Georges Oppenheim, Jean Michel Poggi. Image Processing Toolbox For User with MATLAB. MathWorks,Inc.2001
[21] Mallat S,Zhang S. Characterization of signals from multiscale edges. IEEE Trans Patts Recog Mach Intel, 1992,14(7):710-735
[22] Lo K T, Zhang X D,Feng J,Wang D S. Universal perceptual weighted zerotree coding for image and video compressing. IEEE Proc Vis Image Signal Process. June,2000,147(3):261-265
[23] Sonja Grgic,Mislav Grgic. Performance Analysis of Image Compression Using Wavelets. June,2001,48(3):682-695
[24] Michel Misiti,Yves Misiti. Wavelet Toolbox For Use with MATLAB. MathWorks,Inc.2001
[25] 吴崇明等编著.基于MATLAB的系统分析与设计——小波变换.西安,西安电子科技大学出版社,2000
[26] 胡国生,彩色图象的小波变换编码.自然科学进展,3(1):26-25,1993
[27] 吴成柯.图象通信.西安:西安电子科技大学出版社,1990
[28] Zeimer R E, Peterson R L. Digital Communication and Spread Spectrum Systems. John Wiley &sons, 1985
[29] 王卫,蔡德钧等.一种多分辨率图象混合编码方案.通信学报,1995(2):71-78
[30] Lim J S. Two-dimensional Signal and Image Processing. NJ: Prentice-Hall, Englewood Cliffs, 1990
[31] 毕原杰.图象编码基础.北京:人民邮电出版社,1986
[32] 付组芸编.信息论基础.北京:电子工业出版社,1989
[33] 吴伯修.信息论与编码.南京:东南大学出版社,1991
[34] 李海波.模型图象编码.通信学报,1993,14(2):69-73
[35] 周东,姚庆东.知识基图象编码.通信学报,1993,14(2):27-33
[36] 夏德深,付德胜.现代图象处理与应用.南京:东南大学出版社,1997
[37] 贾世楼.信息论理论基础.哈尔滨:哈尔滨工业出版社,1987
[38] Dony R D, Haykin S. Neural Networks Approaches to Image Compression. Proceeding of The IEEE, Vol. 83, No. 2, Feb. 1995
[39] Tarun Khanna. Foundation of Neural Networks. Addison-Wesley Publication Company, 1990
[40] 阮秋琦.数字图象处理学.北京:电子工业出版社,2001
[41] 何斌,马天予.Visual C++数字图象处理.北京:人民邮电出版社,2001
[42] 赵容椿.数字图象处理导论.西安:西北工业大学出版社,1996
[43] 吴乐男.数据压缩.北京:电子工业出版社,2000
[44] 焦李成.神经网络的应用与实现.西安:西安电子科技大学出版社,1995
[45] 杨建刚.人工神经网络实用教程.杭州:浙江大学出版社,2001
[46] 闻新,周露,王丹力,熊晓英.MATLAB神经网络应用设计.北京:科学出版社,2000
[47] Pratt W K. Digital Image Processing. Wiley-Interscience, New York, 1978
[48] Jayant N S. Adaptive Quantization with A One-word Memory. Bell Sys. Tech. J., 52, 1119-1144, 1973
[49] Ready P J, Spencer D J. Block Adaptive DPCM Transmission of Image. in NTC Conference Record, 2, 1975
[50] Haykin S. Neural Networks:A Comprehensive Foundation. New York :Mxcmillan, 1994
[51] Pao Y-H. Adaptive Pattern Recognition and Neural Networks. Reading MA:Addison-Wesley,1989