Research On One Dimensional DFT Algorithm For Automatic Optical Inspection Of TFT Substrate Surface Defects

With the development of TFT-LCD to large size,thin and high resolution,the probability of a variety of display defects are greatly increased in the manufacturing process.In order to improve the manufacturing quality,automatic optical surface defect detection technology and equipment has become an indispensable means for the panel display manufacturing industry.But it is very difficult to detect the surface defect of TFT substrate in the Front-End-Of-Line Array process for the TFT-LCD.Because the TFT substrate is regularly arranged with horizontal gate lines and vertical data lines,so as to form the background of repeated texture in the surface image,which is a serious interference to the detection of defects.Traditional edge detection and threshold segmentation methods are difficult to be directly applied to the detection of this type of defect.In order to improve the success rate of detecting defects,the texture background must be eliminated without affecting the original characteristics of the defects.In order to improve the performance of automatic optical surface defect detection for TFT substrate in high generation Array process.In this paper,the problem is analyzed and summarized systematically which arise in the automatic optical detection algorithm based on one-dimensional DFT for TFT substrate surface defects.The researches are carried out thoroughly from theoretical to practical in the 1D DFT algorithms,which are in relation to the improvement of the truncation effect,the automatic selection of the parameters and the uneven illumination correction method.And an automatic optical detection algorithm for TFT substrate surface defect based on two-dimensional DFT is proposed.An experimental platform of automatic optical inspection of TFT substrate surface defects is built.Six kinds of automatic optical detection algorithms of TFT substrate surface defects are compared in the experiment.The accuracy rate is to be improved for the automatic optical inspection technology of TFT substrate surface defects.The main research contents and innovations of this paper are as follows:1?A computational model that can characterize the degree of spectrum leakage is proposed.The spectrum leakage of simple periodic signal and ideal TFT substrate line scan image(complex periodic signal)are analyzed in the spectrum analysis.The spectrum leakage of the actual TFT substrate line scan image(approximately complex periodic signal)is discussed in the spectral analysis.The relationship between the spectrum leakage degree and the truncation length is established.A method for improving the truncation effect of one dimensional DFT algorithm is presented.2?The automatic selection methods of two key parameters of the period ?x and neighborhood r is studied.An automatic selection algorithm of the period ?x was designed,which utilized mathematical statistics to process 1D faultless image.By extracting the peak position,sorting,computing spacing and the number of statistics,the ?x is gotten accurately.And the selection algorithm was analyzed and verified by the line images of TFT substrate with various length,resolution and illumination.For the automatic selection of neighborhood r,firstly,each line of the faultless surface image for TFT-LCD panel is preprocessed,then its gray level co-occurrence matrix(GLCM)is calculated,and then the homogeneity property of the GLCM is extracted.Finally,the difference of homogeneity is computed,and the optimal neighbor is that level which the difference shows a maximum value.The images with various length including integer-period,non-integer-period and non-integer-period filled into integer-period were processed in the experiment.From the acquisition of r,it can be seen that the r for non-integer-period filled into integer-period is the same as it for integer-period in the same working environment,and by applying the results into actual defect detection.3 ? The shortcomings of wavelet transform for nonuniformity illumination correction are analyzed in the one-dimensional DFT algorithm.The quadratic curve fitting method and smoothing filtering method for the correction of uneven illumination were put forward successively,and then proposed a smoothing-fitting method.Finally,a smoothing-piecewise-fitting method was presented according to surface image of the actual TFT substrate is longer.Through a large number of experiments,the results show that the smoothing-fitting method can well eliminate the uneven illumination,moreover,can also make up for the shortcomings of wavelet transform method,which can detect the interior of defects and be able to distinguish between bright defects and dark defects.4?The experimental platform of the automatic optical inspection system is built,and the high quality surface images of TFT substrate were collected.The judgment basis of the merits and shortcomings for the surface defect detection algorithm is put forward.For the five kinds of surface defect algorithm of TFT substrate based on the automatic optical detection,detection results of various algorithms are analyzed.The effects of rotation on various algorithms are studied.The detection time of various algorithms is computed.The surface defect detection algorithm for the TFT substrate is determined finally,which is suit to use in on-line automatic optical inspection.