Inspection of defect on LCD panel using local mean algorithm based on similarity (ICCAS 2013)

We introduce a method for detecting defects in TFT-LCD images with periodic patterns. We consider single-patterns with one pattern, and multi-patterns that is classified into a primary pattern region, a secondary pattern region and boundary region. After each region of the patterns is inspected by calculating a new image that removes periodic patterns and highlights defects, we can estimate the boundary region by least squares estimation. Finally we propose a local mean algorithm to inspect the boundary region. The each result of inspection is merged into the final binary image in which the defects are indicated. We focus on increasing speed of simulation to adopt practical system, and results of our methods are promising. We present inspection results on different types of images, the proposed method gives more accurate results than existing methods.