We have tested our method on a wide variety of images. Figure 5
shows eight results generated by our method and a complete gallery
of halftone QR codes could be found in the supplementary material.
The default QR code setting of symbol version 5 and error correction
level H is used for generating all the results. Our algorithm
is computationally efficient and takes less than a second to generate
a result using unoptimized codes on a PC with 3.4GHz CPU
and 8 GB system memory. The performance of our algorithm is dependent
on the size of QR code and the number of data modules
involved in the optimization. The halftone nature makes our results
gain visual quality superior to existing visual QR codes as viewed
from a distance; and our representation model that shows stable image
quality under the variation of QR codes also enables an effortless
extension to combine QR codes with video clips to generate
animated halftone QR codes. Although a simple frame-by-frame
synthesis is employed without taking into account the temporal coherence
between consecutive halftone images, the generated results
are quite satisfactory in terms of visual quality and readability during
the video playback (see supplementary video).