Research On Image Registration Technology In PCB Defect Detection

With the development of the electronics industry,the development of printed circuit boards(PCB)tends to be more and more high-density,and traditional manual inspection can no longer meet the requirements of PCB quality inspection.Therefore,the development of high-efficiency and high-precision automatic optical inspection has become the mainstream of PCB defect inspection.In this dissertation,PCB bare board is used as the research object,and the registration method suitable for PCB image and the defect inspection algorithm based on template matching are discussed.Firstly,the image registration method based on SURF is discussed.SURF is an algorithm with high speed and high accuracy in the point feature registration method.In order to solve the problem that there are too many repeating units on PCB image,which affects the correct matching of point features,an image registration method based on SURF and PCB image coordinate constraint is proposed according to the experience that the angle and translation of each PCB image are small under the limitation of mechanical devices.In this method,the matching accuracy and speed of the fixed number of feature point pairs are improved,and the registration speed is also improved.Secondly,the image registration method based on phase correlation is discussed.In order to solve the problem that the phase-related sub-pixel registration method using one side peak has a large error at the integer translation,a new method is proposed to judge whether to use the two-side peak method to correct the error according to the proportion of the difference between the two sides of the peak to the main peak.In order to solve the efficiency problem in the application of the algorithm,the effects of different decreasing sampling rates on the registration speed and accuracy are discussed.Experimental results show that the maximum error and the average error of the improved algorithm are smaller than the original one-sided peak method.The image of 20 million pixels can be obtained with high accuracy under the condition of five times down sampling,and the speed is about 20 times faster.The algorithm is robust and real-time.Finally,on the basis of image registration,a defect detection method based on template matching was proposed,which used image connected domain and image skeleton endpoint to locate the defect position preliminarily,and used image difference method to screen the final defect position.Experiments show that it can detect two kinds of defects,short circuit and open circuit,and it takes about 12 seconds to process a 20 million pixel image.