Research On Image Matching For IC Packaging Equipment

With the development of information technology, IC (Integrated Circuit) technology hasbeen widely used in the defense, aerospace, communications, and other industries and dailylife. As the key generic technologies for IC packaging equipment, machine vision technologyis gradually maturing along with the development of the IC packaging industry. Because ofearlier research and maturer technology, foreign commercial machine vision software packageoccupied most of the market share, such as MIL, HALCON and Congex, etc. These softwarepackages are easy to use, stable performance, high efficiency, but expensive. Therefore,research vision software package is of great significance with independent intellectualproperty rights. In this paper, the image matching algorithm in the vision software packagewas studied. A gray-based matching algorithm based on Zernike moments and a feature-basedmatching algorithm based on generalized Hough transform are developed to meet therequirements of high-speed, high-precision, rotation-adapted positioning technology in ICpackaging equipment.The image matching algorithm can be classified into two broad categories, thegray-based matching algorithm and feature-based matching algorithm. The gray-basedmethods are robust to noise. Compared with the feature-based methods, it can be applied toblurred image. The feature-based methods are robust to light. Compared with the gray-basedmethods, it can be applied to incomplete image and occlusion image. The gray-based methodsand feature-based methods have their own advantages and disadvantages, we should choosethe appropriate matching algorithm according to the requirements of the application.In the gray-based methods，a fast matching algorithm based on Zernike moments isdeveloped. According to the symmetric of trigonometric functions, a fast approach isproposed to compute the Zernike moments, which gains Zernike moments merely bycalculation of basis function values in the1/8unit circle. Afterwards, the matching point iscalculated by the correlations of multiple moments. Thereby, the sub-pixel matching positionis achieved by Least Square Estimations (LSE). Finally, the rotation angle is estimated bysynthesizing the Zernike moments phase information. Meanwhile, it is used to further speedup matching algorithm which building RCS table to save the relevant information of fastcalculate Zernike moments offline and looking up RCS table online. Experimental resultsshow that the position matching accuracy can reach0.5pixels and the angle matchingaccuracy can be reached more than0.1°，the speed of the fast algorithm is twice or even triplethat of the non-fast algorithm.In the feature-based methods, a rotation adapted matching algorithm based on generalized Hough transform is developed. On the foundation the thoughts of generalizedHough transform, it solves the rotational image matching problem by utilizing the relationshipbetween the gradient direction and the rotation angle. After analysis the voting model,Gaussian voting model is proposed. Meanwhile, the algorithm is designed and implementaccording to the theoretical analysis. Firstly, R-table is built offline. Then, matching positionand angle are obtained by looking up R-table online. Finally, the sub-pixel matching positionand angle is achieved by Least Square Estimations (LSE). Experimental results show that theposition matching accuracy can reach0.5pixels and the angle matching accuracy can bereached more than0.5°，the matching time is165ms. And zero angle matching time is16ms,full angle matching time is70ms.