This dissertation reports on a project that assessed a procedure for recovering defaced serial numbers with marginal human error, based on a semi-automated, nondestructive method using lock-in thermography (LIT) and pattern recognition technique.
One chapter outlines a review of relevant literature and background information for serial number restoration, feature extraction using infrared lock-in thermography, and image analysis. This is followed by a chapter that describes the project’s experimental design, including sample preparation, environment, and control variables to collect thermal imaging data based on the LIT review. A chapter provides results for various samples based on ACIP applied to images generated by LIT analysis. The project concludes that the success of the revised non-destructive method on all samples shows the possibility of its use as a standard method for serial number restoration. The experiments validate the flexibility and consistency of the lock-in procedures across varying defacing, techniques, stamping, and samples. ACIP is a robust method for features pattern recognition to identify the defaced serial number and non-stamped areas regardless of the degree of contrast between them. With fusion classification, the risk of misclassification was minimized, assuring the correct identification of a number or letter. Future work is described. Extensive figures, tables, and references
810 Seventh Street NW, Washington, DC 20531, United States