Spatial filtering with multi-scale segmentation based on gaussian function

Abstract

This research mainly focuses in developing the algorithm for vision inspection inside the TFT-LCD panels. The purpose of this algorithm is to extract the defects precisely from the source images with regular patterns and the falloff of illumination caused by the vignetting effect. Specific spatial masks are generated from the Gaussian function by means of the concept of normal distribution. The source image is decomposed by the Gaussian masks into different spatial frequencies for the scale-space representation. The experimental result shows that the defect image can be effectively segmented from the source image with the smallest size less than 0.5pixel. The optimized algorithm is also provided in this paper for shortening the process time to meet the industrial needs in AOI equipment. This paper provides a novel and complete solution for scale manipulation in image processing. © 2008 Springer Berlin Heidelberg.