Detection of defects in textures with alignment error for real-time line-scan web inspection systems

Hash functions are recently introduced as an extremely efficient method to calculate a measure of deficiency on repeating textures. These hash functions generate one dimensional signatures of patterns and they are simple enough to fit into a medium size FPGA. Although these functions are immune to change in illumination and contrast, they require the texture to be perfectly aligned. Any angular misalignment causes these functions to operate incorrectly. In this paper, a new family of hash function and a new signature analysis method are presented to overcome this problem.