I am looking into the ZXing library source code. I would like as robust as possible scanning of codes that can be printed on bent surface.
From my understanding, ZXing locates the three corners of the QR code, then guesses the position of the fourth, and then reads the code in using basically a perspective grid.
Are there any plans for implementing a more robust method? I was thinking about actually determining the skew of the three corners and then interpolating this information across the code. Another option would be using some sort of neural net to snap onto the black/white transitions, guessing the most likely distortion of the code.