| In forensics, identifying the obliterated serial number unique to each firearm is very important in assisting a criminal investigation. A non-destructive method based on infrared lock-in thermography (LIT)and pattern recognition technique were developed for recovery of the defaced serial with marginal human bias. This thesis focuses on assessing the developed semi-automated method for consistency, validity, and comprehensive automation. Amplitude and phase shift using LIT is suitable to distinguish defective and non-defective regions based on differences in thermal emissivity at an optimallock-infrequency. The technique has a limitation on shallow defects caused by imprint, which leads to low signal to noise ratio. Image analysis, such as smoothing, segmentation, and morphological operations, can enhance feature extraction. Image description technique, such as pseudo-Zernike moments, is used to characterize the extracted feature distinctively. Pattern recognition based on the library of reference alphanumeric characters can be achieved with fusion classification. These techniques are aligned to formulate an automatic character identification protocol(ACIP)for serial number restoration.Keywords:Infrared Lock-in thermography, Serial number restoration, Thermal emissivity,Feature extraction, Lock-infrequency, Pseudo-Zernike moments, Pattern recognition, and Fusion classification |