Research About Key Image Processing Technology In Defect Inspection System Of FPC Based On The AOI

Currently FPC defect detection task is dealt with manually, which costs too muchmanpower and it is difficult to ensure quality of detection. Therefore, new means ofdetection is strongly required to help detection. With the development of imageprocessing technology, nowadays, detection based on computer vision technology hasbeen widely used in the producing and detection of sophisticated electronicmanufacturing equipment and related sessions.The main differences between FPC and traditional PCB are:1. Overall reasonable deformation is allowed for FPC due to its flexible nature,and more detection methods based on rules should be used.2. In FPC, proportion of lines is not high, and that of lumps which is similar toPCB copper pour and play the role of Conductive is higher, so it is necessary todesign specifically detection algorithm for lumps.The research is supported by National Science and Technology Major Project of02special,"863"Program and Science and Information Bureau. Research about keyimage algorithm for FPC defect detection system is introduced in this thesis, whichcontains the following aspect that can meet main requirements of FPC defectdetection:1. Image stitching algorithm for FPC defect detection system. SIFT and SURFimage matching algorithm are employed to generate feature points and match featurepoints, we then exclude mismatch points with the help of camera trajectories of FPCdefect detection system to achieve image stitching task. On the other hand, a saliencybased image matching method is introduced. This method first finds areas with highsaliency index in each image, and then solves the image transformation parameters bymatching these saliency areas to stitch the two images. Furthermore, the method getsenlightenment from low-resolution image and excavates information inhigh-resolution image and can get same result, but its speed is faster. Experimentalresults show that the saliency based image matching method can meet therequirements of speed and accuracy. What’s more, image rotation angle does not much affect this method’s performance.2. Recognition of Mark points and overall transformation of image. Methods ofsearching connective areas and invariant geometric moments are employed torecognize Mark points. After getting Mark points by these two methods, we transformoverall image by using positions of2Mark points. Experimental results show thatrecognition and locating of Mark points can meet the requirements of speed andaccuracy and make preparation for defect detection.3. A detection method based on mixing method is designed. The method involvesdetecting the width of lines, narrowing range of defect target, defect detection oflumps and so on."Narrow connection" division methods are used to narrow range ofdefect target, methods of thinning lines and counting number of pixels in normaldirection are used to get width of lines, and methods of detecting arcs, corner pointsand angles of corner points are used to detect lumps. Experimental results show thatthe methods can detect defect accurately and meet the requirements of speed andaccuracy.