In this paper, we present an automatic algorithm to create a new
type of visual QR code, called halftone QR code, that combines
halftone images with QR codes (see Figure 1 and Figure 2(d)). To
eliminate the aforementioned problems, the key insight is a representation model that minimally binds to original modules and is
flexible to adapt to target halftone images. In our approach, we subdivide each module into 3 × 3 submodules and bind the module’s
color to the center submodule, which leaves the remaining eight
submodules free to independently change their appearance. We use
this flexibility to characterize the appearance of modules using a set
ofbinary patterns. In addition, we evaluate the pattern reliability of
each binary pattern that represents a probability of retaining module’s readability when it is replaced by the pattern. Specifically, we
estimate the probability by learning from a database of synthetic
QR codes that is automatically generated using binary patterns and
a barcode reader is employed to statistically evaluate pattern’s reliability. Finally, to generate halftone QR codes, we introduce a novel
optimization method to compute and assign a binary pattern to each
module by balancing two competing terms, reliability and regularization, corresponding to preferring the assignment of high reliability patterns and regularizing the appearance of modules towards
target halftone images, respectively. (Please refer to the project page
to create your own halftone QR codes for non-commercial use.)