The Research Of Integrated Circuit Chip Image Processing Technique

In this thesis,we research the key problems in IC chip image processing,including edge completion and graphics recognition,template matching and unit recognition,image registration technique.First,according to the characteristic of some IC chip images,which have regular structure,we design a perceptual organization and shape recognition system based on knowledge. This system follows the principle of intelligent,open,hierarchical and specialized,utilizing knowledge furthest. Here,knowledge contains parameters library,shapes library,rules library and human assistance. In this system,we define a pattern description language to describe shapes based on line We also define line's attributes and operations.Next,In order to recognize some complex IC chip images,in which shapes can't be extracted directly,we design and implement a new template matching technique based on parameterized wiener filter. We put forward the PWFM algorithm which utilize the FFT. The proposed approach yields highly localized sharp peaks,with responses at other points are restrained. Using the PWFM,it is easy to locate matching points accurately,and it yields fewer spurious responses than correlation matching. It is robust for almost all the images. Furthermore,we propose a parallel strategy to solve the multiple templates,multiple images' matching.Then,we propose a skew estimation and rectification algorithm,in which wavelet transform is used ingeniously. We get horizontal edges and vertical edges simultaneously,and compute the sum of edge projections' variance in these two directions as no-skew measure. We rotate the image and compute the no-skew measure,when the no-skew measure maximized,the opposite of the rotated angle is the image's skew angle.After making skew rectification,we do the image registration. Based on several hypotheses about image's overlapping area,we deduce that the minimal overlapping area in one image can act as the template,the maximum overlapping area in the other image can act as the searching area,the PWFM is used to establish the correspondence of the template's coordinates in two images,namely obtain the shiftparameters tx and ty. We also provide a multiple images' registration method. Inmultiple images' registration algorithm,the scale differences between images can be eliminated step by step. Multiple images' registration algorithm is accurate and robust. Finaly,the thesis summarizes our work and points out the next research direction.In summary,our work solves the key problems in IC chip image processing,which advance the development of VLSI anti-design software. What we have done provide a good theoretical and practical foundation for future research.