New Findings in Security Printing and Imaging

© New Findings in Security Printing and Imaging Steven Simske, Guy Adams, Jason Aronoff, Margaret Sturgill HP Laboratories HPL-2009-328 security printing, barcode compression, down-sampling, color Tiles: In the past three NIP conferences, we have presented novel techniques for embedding security information into the variable data printing (VDP) for labels, documents and packaging. Recently, we have discovered that fundamental assumptions in other fields error correcting code (ECC), classification and the use of color, for three examples are not necessarily valid when functional security is the goal of the printing. It will be shown herein that the assumptions behind default ECC do not hold for 2D or color 2D barcodes with mobile image capture. Alternatives to ECC are suggested. When security-related tasks are the end goals of the printing and imaging, the deployment optimization for image classification is fundamentally altered. Tuning the classification engine to the security aim, rather than the image features as in traditional classification, provides significant improvement in both image throughput and classification accuracy for security-related tasks such as inspection, authentication and forensic imaging. It will, finally, be shown that there are significant new security approaches either only possible or else greatly enhanced by the printing of color, rather than grayscale or binary, images. These broad results indicate that applied research in security printing requires new basic research in the related image processing fields. External Posting Date: September 21, 2009 [Fulltext] Approved for External Publication Internal Posting Date: September 21, 2009 [Fulltext] To be presented at IST NIP25, Louisville, KY, USA. Copyright IST NIP25. New Findings in Security Printing and Imaging Steven Simske, Guy Adams, Jason Aronoff and Margaret Sturgill, Hewlett-Packard Labs; Fort Collins, CO, USA and Bristol, UK Abstract In the past three NIP conferences, we have presented novel techniques for embedding security information into the variable data printing (VDP) for labels, documents and packaging. Recently, we have discovered that fundamental assumptions in other fields—error correcting code (ECC), classification and the use of color, for three examples—are not necessarily valid when functional security is the goal of the printing. It will be shown herein that the assumptions behind default ECC do not hold for 2D or color 2D barcodes with mobile image capture. Alternatives to ECC are suggested. W hen security-related tasks are the end goals of the printing and imaging, the deployment optimization for image classification is fundamentally altered. Tuning the classification engine to the security aim, rather than the image features as in traditional classification, provides significant improvement in both image throughput and classification accuracy for security-related tasks—such as inspection, authentication and forensic imaging. It will, finally, be shown that there are significant new security approaches either only possible or else greatly enhanced by the printing of color, rather than grayscale or binary, images. These broad results indicate that applied research in security printing requires new basic research in the related image processing fields.In the past three NIP conferences, we have presented novel techniques for embedding security information into the variable data printing (VDP) for labels, documents and packaging. Recently, we have discovered that fundamental assumptions in other fields—error correcting code (ECC), classification and the use of color, for three examples—are not necessarily valid when functional security is the goal of the printing. It will be shown herein that the assumptions behind default ECC do not hold for 2D or color 2D barcodes with mobile image capture. Alternatives to ECC are suggested. W hen security-related tasks are the end goals of the printing and imaging, the deployment optimization for image classification is fundamentally altered. Tuning the classification engine to the security aim, rather than the image features as in traditional classification, provides significant improvement in both image throughput and classification accuracy for security-related tasks—such as inspection, authentication and forensic imaging. It will, finally, be shown that there are significant new security approaches either only possible or else greatly enhanced by the printing of color, rather than grayscale or binary, images. These broad results indicate that applied research in security printing requires new basic research in the related image processing fields. Error Correcting Code Mobile image capture devices are ubiquitous, with increasing ability of consumers, retailers, supply chain managers and manufacturers [1-4] to interrogate products [1,2], labels [1-3], and even signage [4]. Many barcode implementations, however, rely on error-correcting code (ECC), to add robustness to the barcode reading process. The robustness model, however, historically derives from the printing of 1D barcodes; mailing applications; environmental damage associated with smeared ink on low-quality paper; and abrasion or puncture damage. For the reading of barcodes with cameras on mobile devices, localized damage is arguably a less important consideration than overall low image quality during capture. Poor or nonuniform illumination, blur due to poor focus and/or motion, and poor quality printing can cause low quality capture. And ECC is not necessarily designed to overcome these distortions. It has been argued that the selection ofECC, based on communication theory, is a largely misplaced focus, particularly for defects introduced in the process ofprinting or ofscanning [5]. The same reference argues for “ finessing the size of the spots and cells [to] minimize the effects of printing defects”, and continues “ by increasing the size of the data features...virtually any anticipated problem in these domains can be compensated for, so that information can be perfectly communicated” [5]. In this paper, a wide range of values for Error Correction by Percentage of Symbol Area (ECPSA) are compared to uncorrected barcodes with the same density ofbits per unit area. Separately, print-scan (PS), low quality printing and blurring conditions are used to provide test cases for ECC/non-ECC comparisons. A series of barcode readability tests were performed using Aztec symbology high-capacity 2D matrix barcodes. Aztec is able to encode both ASCII and Extended ASCII characters, and when using its full range mode of 151 modules and with 25% ECPSA, Aztec is able to encode up to 3000 characters or 3750 numeric digits (that is, its size ranges between 15 X 15 modules and 177 x 177 modules). All experiments were performed using a 27 x 27 module configuration, so that comparisons between images with the same module size were not affected by overall barcode size. For every test, module size was varied from8 to 30 mils in 1 mil increments (1 mil = 10 in). Each module is either black or white. BCoder Professional software (TAL Technologies, Inc., Version 4.0) allows the varying ofECPSA and payload (this is why Aztec symbology was chosen). Through iterative adjustment of these settings, we were able to obtain 27 x 27 module Aztec barcodes with 0%, 10%, 20%, 30%, 40%, and 50% ECPSA settings (Table 1). The number of payload modules (or “ bits”), P M , was equal to 648, 584, 520, 456, 392, and 328, respectively, for these ECPSA settings, so that the number of ECPSA, or non-payload modules, NP M , was equal to 0, 64, 128, 192, 256and 320 bits, respectively. For all of the following tests, the barcodes were read using an InData Systems 9500LDS portable terminal with add-on optic “ shroud” for 405-nm LED (light-emitting diode) Light Delivery System (LDS-V2), hereafter “ IDS-LDS”. This system provides uniform lighting conditions (405 nmillumination) for all barcode reading performed (so that we are sure consistent illumination was used throughout the experiments). Multiple pages (20 or more barcodes at each of the 23 module sizes) were printed under the following experimental conditions: