Printed Circuit Board Automatic Optical Inspection Of The Registration Algorithm

In this dissertation,the image registration algorithms used in automatic optical inspection(AOI) of printed circuit board(PCB) are investigated.For AOI technology,imaging system is used to grab the digital images of PCBs,and then these images are processed with image analysis toolkits to find the defected parts on them.Evaluations about the quality of these PCBs will be sent to manufacturing company at last to helping them with quality control process(QC process).The digital image of the PCB is needed to be compared with a reference image of zero-fault PCB in the inspection process.In order to accomplish comparison,these two images must be aligned in space precisely at first. This is the mission of the image registration algorithm.Image registration is a crucial step in AOI system.It determines the accuracy and the spatial resolution of the inspection system.According to the working flow of image registration methods, registration process is usually split into five steps,which are preprocessing,feature extraction,feature matching,transformation model estimation and image resampling in sequence.Details about these five steps are discussed in this dissertation.In the step of preprocessing,a fast algorithm to implement median filtering is proposed.This is a smart method using moving histogram technology to speed up the computing procedure dramatically.In the same time,an image restoration method is proposed to overcome the uneven illuminating situation.In this method,wavelet tools are used to extract the main tendency of the uneven brightness.Data and graphics of the result are list after the description of them.In the step of feature extraction,point features,line pattern features,region features are all mentioned.A corner detection algorithm is proposed to find point features;A Canny-like edge detector for color image is given to get the contour features;A Bayesian classifier is also designed to classify the pixels in the PCB image to extraction all the small components on board.In the step of feature matching,matching algorithm based on Hausdorff distance is described in detail.A fast method to compute the Hausdorff distance is also proposed accompanied with some experiment data.Feature matching experiments using invariant moments of shapes are discussed meanwhile.In the last two steps of the registration process,the reason for missing alignment of the PCB images has been analyzed.With the conclusion of the analysis,a suitable transformation model has been chosen to describe the missing of alignment.Furthermore,optimization techniques for computer programming are list in chapter 5.The efficiency of the codes is an essential problem in registration algorithm when used in industry situations to fulfill the demands of real-time processing.The innovation of this dissertation is the edge detector and Bayesian classifier using in the step of feature extraction,both of them is aiming at color images.They are all new ideas in AOI system for PCB.AOI systems that use color images are not available until recent years in West and in Japan.In our mainland,there are still no reports on it.Methods mentioned in this paper are almost used in projects under way,and they will play an important role in the success of developing our AOI system.