For a university honours project, I am developing an algorithm to read the text on nutrition labels.
One difficulty I have encountered is how to 'flatten' the text in images of labels that are very deformed, such as the one on this ball of mozzarella:
A lot of research has been done on shape estimation of printed documents using generalised cylindrical models, but I don't really think that model is applicable here. It would also not work for images of nutrition labels on chip packets, for example.
However, by virtue of the rectangular nature of the label (when originally printed), I had the idea that it might be possible to approximately estimate the deformation by looking at the borders and lines in the table. I am willing to assume the presence of such lines in any image I need to flatten.
Can anyone point me to some relevant research or give any other advice?