| In recent years,the development of the machine vision industry is fast in our country.In the process of product manufacture,it is unavoidable to encounter various damage and defect in the surface.One of the most difficult to detect is a kind of surface defects with low contrast,edge blur,different shape and different size.The traditional manual detection methods cannot satisfy the needs of modern production,for the questions exiting in it such as high cost,low efficiency and high error detection ratio.So,it is urgent demand to detect and segment the low-contrast surface defect use the automated method with stability,precision and high speed.This research topic is aim to solve the difficult question to inspect the low-contrast surface defect of industrial product.After investigating the effect of image segmentation method,image background reconstruction method,characteristics classification and other methods for detecting the low-contrast surface defect,the background reconstruction method is assigned as as the main research direction.First,we analyze several classical detection methods for surface defect based on background reconstruction.Due to the detect result of the low-contrast surface defect detection method is not accurate and serious noise pollution,the idea is put forward based on image background removal and highlight the defect detection.The steps are as follows:First,the method with the sample learning is used to construct a low noise,low loss of high quality background model and candidate defects contour is obtained by removing image background;Secondly,gray stretch,median filtering and dilation operation are carried out on difference image after background suppression to enhance the visibility of defect area;Finally,accurate positioning and segmentation is performed for target defect area.The article mainly works on the following four aspects,including the main innovation points.● Put forward the background reconstruction method based on sample learning and image bases.Based on the independent component analysis model,the mechanism of sample learning and image representation is introduced to reconstruct the background image.The reconstruction background image can keep background information as much as possible.● Put forward the gray level stretch method based on mean and variance of difference image,which can both enhance the potential defect area and reduce the noise interference.● Apply the level set method based on the improved CV model to locate and segment the defect area accurately from the difference image.The adaptive internal and external gray differences item and penalties item are introduced to the CV model,which can avoid the noise interference and make the results more realistic.Finally,we set up a system platform for low-contrast surface defect detection,and applied the platform to the TFT-LCD mura defect detection.The detect results shows that the proposed approach has a certain application prospect comparing with the traditional detection method. |
