It would be helpful to get more details on the "meaning" of the imaging model, equation (2). I think I get the model for m_hat (eqn (1)): basically, change the illumination abit, transform the image, convolve with the PSF to model the optics, then down-sample.

But I am not clear on how we go from that, to the equation for the nth image (eqn (2)).

Sorry for the late question . If you don't get to it, I won't mind.

This seems like a nice complement to the text reading paper you presented earlier. For your license plate reading project, it seems like both are necessary. I was wondering if you plan to use the two as exclusive components, or if you can somehow combine the two methods to obtain improved results?

What was the basis for using the Huber function as the prior for MAP? Is that something that can be derived from basic principles, or was it choosen simply because it seems to work for this? Is there an analytic means for setting its parameter values, or is that something you just have to tweak for every application (and if so, how narrow a range is it effective for, i.e., how brittle is it)?

Absolutely. I've prepared a few slides on point spread functions, so hopefully that will clear things up. I'll try to answer your other questions as well.

Louka, can you explain point spread functions in detail? The paper does not do this, yet it depends highly on these functions.

Engineering-wise, are we using a common region of the low resolution frames to get the super-resolution image? Otherwise, some frames may not contain the whole area of the image that we are interested in. But yet then what would be different between the frames?

Are there any conditions where if we select a subset of all the frames, we perform better?