With the development of high-generation production lines for Liquid Crystal Display (LCD) panel, online real-time detection of large-scale, high-resolution LCD surface defect imposes stringent requirements on image algorithms and computing platforms. This paper aims at the edge effect in the traditional one-dimensional Fourier reconstruction LCD detection algorithm. Based on the potential periodicity of the Fourier transform and the approximate periodic arrangement of the element array in LCD panel images, a complete period truncation image preprocessing strategy is proposed. The processing strategy not only eliminates the ring effect at the connection joint caused by the original algorithm's extending period at both ends of the image, but also reduces the length of the Fourier operation, which is even more noticeable when detecting large-size panels. In addition, in order to adapt to the fast Fourier transform input interface, the resampling method is applied to the complete period truncated image whose length is not an integer power of 2, and an FPGA-based resampling acceleration structure is designed. The experimental results show that compared with the traditional one-dimensional Fourier reconstruction algorithm, our proposed method not only performs better in eliminating LCD global periodic texture, but also reduces the amount of data to be processed. The FPGA acceleration scheme even reduces the scanning detection time of a 8.5-generation LCD panel to 12.5s

Date Published: 7 March 2019
PDF: 6 pages
Proc. SPIE 11053, Tenth International Symposium on Precision Engineering Measurements and Instrumentation, 110532Z (7 March 2019); doi: 10.1117/12.2512075

Published in SPIE Proceedings Vol. 11053:
Tenth International Symposium on Precision Engineering Measurements and Instrumentation
Jiubin Tan; Jie Lin, Editor(s)