Research On The Automatic Detection Of TFT-LCD Mura Defects

TFT-LCD Mura defect is a common surface defect which is different from the conventional one, it has the characteristics of low contrast, variety of shapes and unequal brightness, the common manual detection methods exist omission, fault detection problems. In contrast, the Mura defect detection based on machine vision is popular for its low cost and high output. At present, the research on this method is mainly focused on how to improve the precision of detection.By analyzing the difficulties in the detection of mura defects, and comparing with the existing detection methods, This thesis proposes a detection method based on separating spatial domain reconstruction background, and studies the key technologies such as brightness noise improvement, defect segmentation and so on. The main work is as follows:(1) Separating background and highlighting mura defects. This thesis derives the calculation formula of singular value and image energy by matrix model based on the distribution of the singular value matrix of image pixels, designs and derive the calculation relationship between entropy and the singular values with the help of energy and information affine, further to use the maximum entropy probability distribution theoretical derivation background component singular value distribution. Considering the deviation of the reconstruction image produced by the truncated singular value after the initial background model reconstruction, this thesis also studies the compensation factor ? which is restricted by low noise, low loss and high brightness three quality control mechanism, in order to further optimize the quality of the background model and to realize the effective separation of the background model.(2) Reserching on improvement of defect image after background suppression. The mura defect after background suppression has the characteristics of uneven and low brightness, before the segmentation process, it is necessary to enhance its SNR. We reach a conclusion that the noise is mainly distributed in the gray near to zero by analyzing the background suppression composition and image after histogram, then design a simple and fast method to enhance the defect area brightness using linear gray transform. In the next, we eliminate the noise by morphological image denoising, mainly through the combination of corrosion and expansion operations to maintain the brightness of target area and reduce noise.(3) Mura defect segmentation optimization. Taking into account of a certain gap between the theoretical value and the actual value, the mura defect image will still be contaminated by noise, we put forward a improved C-V model segmentation method. The B-spline continuous level set function is used to replace the traditional one to describe the boundary, so as to improve the anti-noise performance. And through the whole divergence to eliminating the dependence of the difference measurement on the coordinate system, reduce the impact of uneven brightness on segmentation results.Finally, according to the research results, the automatic detection simulation system of LCD-Mura defects is established, and the samples are tested and verified in the laboratory environment. The quantitative results show that the method of this thesis has better detection precision and accuracy, and the detection rate can reach 96 percent, compared with the traditional method, has a certain application prospect.