使用參考比對法之印刷電路板底片缺陷檢測

• Main Authors: Yu-Kang Chang, 張育康
• Other Authors: Yih-Chih Chiou
• Format: Others
• Language: zh-TW
• Published: 2003
• Online Access: http://ndltd.ncl.edu.tw/handle/18562323157705452500

碩士 === 中華大學 === 機械與航太工程研究所 === 91 === We have developed an Automatic Optical Inspection system for detecting defects on a PCB film. The detection algorithm is based on the logical operations on two images, known as image subtraction. That is, by subtracting the standard image from the test image, the resulting image will show the difference between the two images. If the result is a total black image, it means that the two images are exactly the same. Otherwise, the contents of the resulting image will represent the defects. To guarantee the success of subtraction-based defect detection method, it needs to overcome the problems of illumination variance and image misregistration. With respect to the two problems, we use histogram equalization technique, especially the hyperbolic cube root type, to solve the problem of illumination variance. And then, we use a novel feature-corresponding technique to solve the problem of image misregistration. First, features are extracted from the test image and the standard image separately. Second, according to the moment invariant characteristic, the matched feature pairs can be found. Third, by assuming that the two images are differed only by affine transformation, then based on the matched pairs, the parameters of the coordinate transformation can be found. Finally, the two images can be registered by applying the coordinate transformation to each point in the test image. 200 pairs of film images have been used to test the performance of the proposed subtraction-based defect detection system. The test results show that only 11 out of the 200 images are failed, therefore the detection rate is 94.5%. The registration accuracy evaluated by the root mean square error is about 0.8 pixels on average. As to the speed, it takes 47.76 seconds in total to complete the inspection of the 200 images; therefore the detection speed is 0.239 second per image on average.