QQ客服 官方微博 反馈留言

题名	光学图像处理中几个问题的研究
作者	陈戊
学位类别	博士后
答辩日期	2005
授予单位	中国科学院上海光学精密机械研究所
导师	徐至展
关键词	神经网络 误差 衍射 空间光调制器 图像形态学 模式匹配 文档图像压缩 字体识别
其他题名	STUDY ON SOME PROBLEMS OF OPTICAL IMAGE PROCESSING
中文摘要	本文主要分为4部分。1、光电混合神经网络系统衍射误差的仿真研究。 本文介绍了3种类型的光电混合神经网络系统，重点研究了衍射引起的误差对系统输出的影响。根据衍射、光学信,r}、处理和神经网络理论，采用实际实验中的参数和输入数据，对衍射造成的输出误差做了仿真分析。分析表明，瑞利一索末菲区衍射造成较大的输出误差。瑞利一索末菲区与远场衍射综合作用时，误差因输入图像模式不同而差异较大;其中，当输入较小的简单图像时相对误差较大;而对实验中实际采用的复杂图像，相对误差较小。利用线性回归方法对输出数据做了校正，并分析了其可行性，校正后的数据误差降低一个量级。衍射误差对实验中识别率的影响可以控制在较小的范围内，识别率可以保持在97.7%以上。2、基于图像形态学和模式匹配的中文文档图像压缩。本文提出了一个高效的、用于中文文档图像的压缩方案，这个方案是基于图像形态学和模式匹配的。其中使用了图像形态学操作方法对汉字进行分解和重组。在重组过程中使用3个判据，特别强调了连通性的重要性，并以此为限定条件构建“最小误差位图”。在模式匹配过程中，设计了适合汉字字型特点的匹配判据。在大尺度模式的空白区域被用来嵌入小尺度模式，从而减小了模式库的尺度。算术编码应用于最后步骤的压缩。本文的方法获得了明显高于比其他各种方法的压缩比，并能保证在内容无损基础上的图像重建。
英文摘要	This thesis consists of 4 parts. 1、Diffraction error numerical analysis on an opto-electric hybrid neural networks. In this thesis, three types of opto-electronic hybrid neural networks are introduced. The errors induced by diffraction in these networks are analyzed with computer simulation where diffraction, optical information processing and neural network theories are applied. Rayleigh-Sommerfeld diffraction induces large magnitudes of relative errors. When Rayleigh-Sommerfeld diffraction and far field diffraction are employed together, the conclusions are different according to the display modes. When the display mode is small image, the magnitudes of relative errors are very large. However, when the display mode is complicated image, the magnitudes of relative errors are small. The feasibility of using linear regression to calibrate the output data is discussed in this paper. We find that linear regression can reduce the errors for about one magnitude. According to the analyses, we conclude that the errors induced by diffraction can be minimized to a low level in the experiments, and thereby, the recognition rates can be remained at a high level (larger than 97.7%). 2. Chinese ocument Image Compression Based on Morphologic Analysis and Pattern Matching. We propose a highly efficient content-lossless compression scheme for Chinese document images. The scheme combines morphologic analysis and pattern matching to cluster patterns. The morphologic analysis is applied to decomposing and recomposing the Chinese character patterns in order to achieve the error maps with minimal error numbers. In pattern matching, the criteria are designed to be adaptive to the characteristic of Chinese characters. The blank spaces of large-size patterns may be embedded by small-size components in order to achieve small-size pattern library image. Arithmetic coding is applied to the final compression. Our method achieves much better compression performance than most of alternative methods, and can assure content-lossless reconstruction.
语种	中文
内容类型	学位论文
源URL	[http://ir.siom.ac.cn/handle/181231/16026]
专题	上海光学精密机械研究所_学位论文
推荐引用方式 GB/T 7714	陈戊. 光学图像处理中几个问题的研究[D]. 中国科学院上海光学精密机械研究所. 2005.

个性服务

查看访问统计

相关权益政策

暂无数据

收藏/分享

除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。