Defect detection in ID cards with accurately reconstructed reference image

Abstract

ID card is made by hot-pressing a standard film with identifiable information onto a fixed baseboard with background of wavy lines. In this paper, we propose a defect detect algorithm by synthesising the film image and baseboard image to accurately reconstruct a reference image for a test card. First, to ensure the content consistency in position and scale, we align the card to a standard film image through perspective transformation(PT) based on AKAZE key-points. Besides, we use contrast limited adaptive histogram equalization(CLAHE) to enhance the background pattern of a baseboard image, and then align it to the rectified card. Second, we apply multiply algorithm to synthesise the aligned film image and baseboard image as a reconstructed reference image. Besides, we align the lightness histogram of the reference image to a test card so as to eliminate the lighting difference. Finally, we apply the difference image method based on canny edge detection to detect difference between a reference image with a card, and further extract the defect information. We experiment on cards with different types of defects and shooting disturbances. Results show high accuracy of our method.