CASIA OpenIR  > 毕业生  > 硕士学位论文
集成电路图像处理相关问题研究
Alternative TitleResearch For Integrated Circuit Image Processing related Problems
孙志宏
Subtype工学硕士
Thesis Advisor彭思龙
2004-06-14
Degree Grantor中国科学院研究生院
Place of Conferral中国科学院自动化研究所
Degree Discipline模式识别与智能系统
Keyword集成电路图像 Lvq 线结构特征 Snake 线目标 Integrate Circuit Image Processing Lvq Line Structure Character Snake Object Lines
Abstract集成电路图像处理就是利用图像处理、模式识别等理论和方法,对用光学 显微镜或者电子显微镜采集到的芯片图像进行处理,以识别出上层图像实例间 的连接关系以及底层图像中的单元区。根据图像处理后得到的数据,来进行原 理图的识别与提取、版图提取、电路综合理解。这整个过程构成集成电路反向 分析系统。因此,经过图像处理得到的数据的准确性直接影响到集成电路反向 分析的结果的正确与否,而集成电路反向分析对于提高我国的芯片设计水平和国 防安全有着重要意义。 集成电路图像处理分为两个部分:一是线目标的提取与识别,即上层图像 实例间连接关系的识别;二是底层单元区识别。集成电路图像通过解剖芯片, 用专门的图像采集系统采集得到,对于不同的芯片,图像呈现出不同的特征, 但对同一芯片的一批图像,具有某些一致性,比如质量比较好的图像、背景和 线目标灰度值差距比较大,同一层图像线结构比较简单,同为水平线或者竖直 线。因此,我们可以通过选择合适的特征,如数学特征,结构特征,来反映整 批图像的一致性属性。 本文在线目标提取部分是对质量比较差的集成电路图像进行处理。该类图 像的特点是:本来的连续线由于腐蚀或成像原因而断开,并且某些背景区的灰 度值高于线目标的灰度值,即此类图像可以分为三类:暗背景区,亮背景区和 线目标区。线目标提取问题可描述为:将图像分割为背景区和线目标区,并且 连接断线。我们选取L,VO网络,先选择合适的结构特征作为LVQ的输入,然后 通过神经网络来学习用数学表达式难以表达的断线规律,同时通过训练,将暗 背景,亮背景,线目标量化成不同的矢量(权重),然后将此矢量作用于整批图 像完成图像分割。对于单元区的识别,用snake的方法来得到单元器件的形状, 为后续识别做准备,并提出了一种改进的snake模型,来解决原有的snake不能 收敛到深的凹腔的问题。
Other AbstractIntegrate circuit (IC) image processing deals with integrate circuit images gotten from optics microscope or electron microscope to obtain the lines between cells in upper layers and cells in lower layer using theories and methods of image processing and pattern recognition. Subsequently, we use the data gotten above to recognize and extract principal map, get the chart and comprehend the logical circuit, which compose the integrate circuit reverse analysis system wholly. So the correctness of the data after image processing directly influences the result of the integrate circuit reverse analysis, which has great significance for improving the design level of CMOS chip and national defense security IC image processing contains two parts: one is lines extraction, i.e. recognition of lines between cells in upper layers, the other is recognition of cells in lower layer. To different CMOS chip, IC images obtained take on different characters. But for the right chip, the whole set of IC images have some consistent attributes. For example, the difference of gray level between background and object is high and the line is horizontal or vertical in the same layer for good quality images. So the proper characters such as mathematical character, structure character can reflect the consistent property of the whole set images. In this paper, the part of lines extraction is to deal with the poor quality images. They have the character as follows: lines which should be continue break for the reason of erosion and imaging; gray level in some background regions is higher than that of object lines, that is, such kind of images can classify into dark background regions, bright background regions and object lines. Then the problem of extraction lines can describe as: segment the image into background regions and lines, and connect the break lines. We choose LVQ neural network as the method. Firstly, we choose proper structure character as the input of LVQ, then we get the rule of break lines which cannot be expressed by mathematic and three different vectors (weights) corresponding to dark background regions, bright background regions and object
shelfnumXWLW791
Other Identifier791
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/6747
Collection毕业生_硕士学位论文
Recommended Citation
GB/T 7714
孙志宏. 集成电路图像处理相关问题研究[D]. 中国科学院自动化研究所. 中国科学院研究生院,2004.
Files in This Item:
File Name/Size DocType Version Access License
硕士生学位论文-791.pdf(10972KB) 暂不开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[孙志宏]'s Articles
Baidu academic
Similar articles in Baidu academic
[孙志宏]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[孙志宏]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.