Mobile Phone Screen Bad Point Detection System Based On Machine Vision Research

With the increasingly competition, the traditional method of manual inspection ofthe quality of mobile phone screen has been unable to adapt to the productionrequirements. So, the domestic mobile phones manufacturers are working on developautomatic inspection system based on machine vision to improve production efficiencyand reduce production costs. Therefore, research and development of a phone screendefect detection system based on machine vision has tremendous market value.Based on the above background, this thesis analyses some researches and cases onindustrial inspection using machine vision in recent years, studies in-depth theknowledge of machine vision, including image acquisition, camera calibration andimage processing technology, designs and implements the defect of dead pixel on cellphone screen detection system based on machine vision.This defect detection system consists of image acquisition module, cameracalibration module and the defect detection module. As the basis of the system, imageacquisition module acquires the image using the technology of DirectShow. Cameracalibration module models camera model, gets the intrinsic matrix and the distortioncoefficient of the Camera, corrects the image distortion to improve the accuracy offeature extraction in the defect detection module. Defect detection module, as the coreof the system, firstly, sets Log Gabor filter bank parameter adaptively using imagepeaks coming from the Converging Squares Algorithm and extracts the edge of imagebased on the phase congruency. Secondly, based on the edge, it segments the region ofdead pixel on cell phone screen using the algorithm of region growing. Finally, it willevaluate the quality of mobile phone screen according to the value of characteristicparameter about the dead pixel defect region.The result of system test in laboratory shows that the cell phone screen dead pixeldefect detection system can evaluate the quality of mobile phone screen accurately inreal time.