Our idea is based on the relationship of equation (6).
By estimating c and d for several parts of an image, we
can see whether the noise in the image is consistent with
the model. Taking images in a video sequence and processing them in pairs as we saw should allow this. However, instead of coming from uniform images captured
under different levels of illumination, the data that we
use comes from locally uniform patches in the images.
As long as there are enough of these patches covering
a minimum range of intensities we should be able to
verify the linear relation given by equation (6). Furthermore, the uniform patches should be large enough to
provide enough information for a good estimate of