An Improved Similarity Measure for Automatic Print Inspection
Barkol, Omer; Kogan, Hadas; Shaked, Doron; Fischer, Mani
Keyword(s): Similarity measure; SSIM; print inspection; subpixel misregistration
Abstract: Motivated by the goal to improve our automatic inspection system for print defect detection that utilizes an in-line scanner [1, 2], we review the SSIM structural similarity measure . We propose two improvements of SSIM for the two following issues: (1) SSIM does not respect linear combinations. That is, patch x might be considered more similar to patch y than to z = ax + (1 ? a)y, a 2 [0, 1]. (2) SSIM does not handle well subpixel misregistration, which is inherent in print inspection, and is invisible for the human eye. We propose a modified version of SSIM that respects linear combinations, and then use this measure to define SubPixel misregistration aware SSIM (SPSSIM).We show that the modified measure performs better in terms of lowering the miss-detections vs. false alarms rate.
Additional Publication Information: Published in ICIP 2010: 17th IEEE International Conference on Image Processing Proceedings
External Posting Date: July 21, 2012 [Fulltext]. Approved for External Publication
Internal Posting Date: July 21, 2012 [Fulltext]