With the prevalence of better sensing capability and computational power in consumer electronic devices, using cameras for communication purposes draws increasing attention from both academia and industry. Extensive works have been done to design efficient printer-camera communication systems, including various famous barcode systems. However, such a print-and-capture communication channel is complicated due to the sophisticated printing and imaging processes involved, which introduces difficulties in modeling the channel and limits the development of the system. In addition, even with a given model, the allowable overhead for training the model is severely limited by the high data capacity requirement and the available computational resources in real-world implementation platfo...[ Read more ]

With the prevalence of better sensing capability and computational power in consumer electronic devices, using cameras for communication purposes draws increasing attention from both academia and industry. Extensive works have been done to design efficient printer-camera communication systems, including various famous barcode systems. However, such a print-and-capture communication channel is complicated due to the sophisticated printing and imaging processes involved, which introduces difficulties in modeling the channel and limits the development of the system. In addition, even with a given model, the allowable overhead for training the model is severely limited by the high data capacity requirement and the available computational resources in real-world implementation platforms. If a model is accurate but too complicated, it is difficult to be applied in practice and is thus less valuable. On the other hand, if a model is over-simplistic, it could be inaccurate and may not give satisfactory performance in practical scenarios.

In this work, we proposed a new model for the print-and-capture channel, which enables low-overhead training to achieve good accuracy. We first derive a new print-and-capture channel model with an explicit parametric form according to the relevant literature. Based on the model, an efficient scheme for training the model and equalizing the channel is proposed. To reduce the training overhead, we decompose the channel distortions into three types by utilizing the structure of the proposed parametric form and conduct a sequence of simple estimation steps instead of tackling the whole channel all at once. Compared with the previous work, the proposed scheme requires less training and computation overhead, making it feasible for practical applications. Furthermore, the proposed model is capable of generating the probabilistic information as the demodulation result, which allows the usage of advanced channel codes with soft decision decoding and thus improves the system performance.

Finally, a multilevel barcode system based on the proposed channel model is designed for achieving higher data capacity. For applications as barcodes, the proposed scheme requires quite few training symbols (less than 5% of all modules) in one barcode. To utilize the probabilistic outputs of demodulation, a low-density parity-check (LDPC) code with the sum-product algorithm (SPA) decoder is applied. Experimental results are presented to verify the effectiveness of the proposed model and the superior performance of the multilevel barcode system under realistic scenarios.