Research Of Air Display Components Lcd Defect Detection Based On Machine Vision

The LCD（Liquid Crystal Display）panel is an important component of aerial display unit. With therapid development of the industry, the quality detection in the production line is crucial. Due to thecomplexity of the manufacturing process and a variety of environment factors, it’s hard to avoid somekinds of the defects while making the LCD. Currently, the defect detection work relies mainly onmanpower, thus, we can only take the method of sampling inspection, the precision and efficiency ofthe detection are low. The research of automatic detection of defects is important for improving thequality of LCD.Firstly, the experimental platform based on the computer vision for the LCD defect detection isintroduced in the paper, and then combined with the knowledge of image processing, the images ofthe LCD panel are tested. Because of the poor contrast of the defect area and the complexity of theLCD background texture, general image segmentation is difficult to extract defects. Both the LCDwith the texture and the one without texture are experimently studied. According to the internalstructure of the LCD panel and the display characteristics of the defects, we hold analysis andexperimental verification for the algorithms from the view of how to suppress the impact ofbackground texture and how to build a background image without defects and how to split theoutstanding defects from the image characteristics respectively. The main findings are listed asfollows:(1) To detect the defects of the LCD without texture, three methods are used and studied, one isbased on the discrete cosine transform, another is based on the two-dimensional surfacereconstruction, the other is residuals calculation based on regression analysis. The first two methodsreconstruct the background image firstly, and then obtain the difference image of the reconstructedimage and the grabbing image, finally set threshold on the difference image to detect the defects. Thethird method first get the residuals of the grabbing image and estimated image, then the thresholdingoperation is ued to detect the defects. In the residual method, the image is splitted into sub-blocks toimprove the calculation speed. The online threshold adjustment method is proposed in this paper tosolve the problem of the threshold manually set.(2) To detect the defects of the LCD with uniform texture, this paper first studies the algorithmsfrom multi angles to remove the background texture. The three methods used are separately based onthe singular value decomposition, Fourier transform and Gabor transform. Finally, a threshold is madeon the image to detect the defects. In the method based on the fourier transform, the selection of the cutoff frequency of the low-pass filter is given according to the distribution of the image textureinformation. This method is also tested on a craced plate and a good test result is obtained. In themethod based on the Gabor transform, the Gabor filter parameters are set based on the characteristicsof the detected LCD and a good test result is got.(3) The three-dimensional Fourier transform is used to detect the defects of the LCD with randomtexture. To remove the random texture, the three-dimensional Fourier transform is imposed on thecontinuous grabbing images and a filter method is designed under the three-dimension.(4) In this paper, several engineering application of the above algorithm is validated. Thealgorithms are experimentally compared from the spects of the effects of gray value difference、time-consuming、threshold selection and the direction of the image. Finally, the algorithm meet theengineering requirement of the project is got.(5) A platform for LCD detection is built. The platform can detect defects for several samples andachive real-time detection.