Accurate Modelling and Efficient Estimation of the Print-Capture Channel with Application in Barcoding.

With the rapid advancement of sensing capability and computational power in consumer electronic devices, the use of mobile cameras for communications has attracted significant attention from both academia and industry. The design of a reliable communication system over the print-capture channel is a key challenge in some important applications, such as barcoding and document authentication. However, the real-world printcapture channel is spatially non-linear and non-stationary, and lacks a standard parametric model. As a consequence, advanced channel coding techniques developed for common channel models are not applicable in the existing systems without costly and device-dependent pre-training. In this work, an accurate parametric print-capture channel model and an efficient channel estimation scheme requiring low training overhead are proposed. With the estimated parameters, the proposed print-capture channel model can be linearized to the (possibly input-dependent) additive white Gaussian noise (AWGN) channel model. This allows the use of state-of-the-art channel coding schemes, such as low-density parity-check (LDPC) codes with iterative soft decoding, to improve the reliability of the communication system under challenging conditions, e.g., at a low signal-to-noise ratio. As an example application of the proposed print-capture channel model, a demonstrative multilevel 2D barcode using an LDPC code with iterative soft decoding, is designed to enhance the reliability over the conventional QR code which is based on the Reed-Solomon code with hard decoding. At a printing resolution of 600dpi, the 8-level 2D barcode achieves data capacity gains of about 10% and 56% over the QR code in the in-focus and blurlimited scenarios, respectively. It is noteworthy that the enhanced data capacity of the multilevel barcode is made possible by the proposed channel estimation scheme, which incurs an additional training overhead of about 3.3% only.