Research On STN-LCD Surface Defect Inspection System Based On Machine Vision

Liquid crystal displays are important devices for good human-computer interaction and are widely used due to their low power consumption.The quality of the display is an important indicator,so the detection of display defects has received more and more attention from the production department.In recent years,due to t he rapid development of computer technology and machine vision,industrial inspection based on image processing has become more and more prominent,such as high efficiency,low cost,and strong stability.Due to the problems of uneven display brightness,s mall defect area,and low contrast of the liquid crystal display,it is difficult to perform automatic detection based on image processing.Traditional detection of liquid crystal display screens is mainly based on template matching alignment,and then cal culation of the statistical information extraction defects such as the average value or variance of each area in the image,and this detection method takes a long time,low real-time,has strict standards for the location of the image.Aiming at the shortcomings of traditional detection methods,this paper proposes a local CV model method to extract defects,describes the shape of defects through geometric moments,and classifies them by classification algorithms.The main research work and innovations in this paper are summarized as follows:(1)The defect area extraction method is studied.For the low defect contrast and small defect area,this paper adopts the local CV model method to extract the defect area.This method firstly extracts the candidate defect area by the fuzzy difference method.That is,the image is blurred and subtracted from the original image.This method can effectively extract the candidate defect area and use it for subsequent processing.Then,the CV model based on the level set is used to find the outline of the candidate defect area.After a certain number of iterations,if the outline is clearly visible,it is a true defect;otherwise,and if the outline disappears,it is a false defect.(2)The defects are classified into point defects and line defects for the types of defects.The shape is an important characteristic of object difference.For the feature of the surface defect shape,the defect region is described by geometric moments,and 7 invariant moments with translation,rotation,and scaling invariance are taken as feature vectors,and the k-nearest neighbor classification is used to classify point and line defects.This method effectively describes the shape features of defect areas and improves the accuracy of defect classification.(3)Calibrate the camera and calculate the size of each pixel to measure the size of the defect.According to the actual needs of liquid crystal display detection,a liquid crystal display detection system is designed and established.The system has the functions of defect extraction,defect classification and defect size measurement,which can effectively solve problems such as long time consuming and low real-time performance of traditional detection methods.The innovation of this paper is to use local CV model method to extract the defect area.This method can effectively avoid the uneven distribution of background brightness,and can further effectively extract defects based on the neighboring pixels of the defect;In the process of categorization of point defects and line defects,the shape of the defect is mainly described by the geometric moment invariant according to the shape of the defect,and the defect is classified as a feature vector.