У меня есть пары изображений. Грубо говоря, одно изображение является размытым изображением другого. Я могу пространственно зарегистрировать эти изображения, но тогда я не уверен, что делать (в частности, учитывая грубую часть).
Действительно, я надеюсь, что кто-то здесь говорит: «Очевидно, используйте алгоритм». Спрашивать не повредит. Заранее спасибо
1) Blind deconvolve that would automatically optimally sharpen the
image and return the filter that does it. The inverse of that
filter would go the other way. I guess you could do that on both
images and compare the filters.
2) divide the FFT of one image by the other to see what filter would
pop out but I suspect doing that naively (considering one image
isn't just an exact blurring of the other) would probably not work
well
3) take the fft of both to get a power distribution (a 1d would
probably do) just to get a rough idea: maybe visual if nothing
else.Registering the images isn't required.
4) use some image quality metric. Registering the images isn't
required. Those metrics are relative metrics of sorts; a yardstick
were the "yard" isn't an exact thing (like a returned convolution
kernel)
5) I could do an optimization on a psf(given a pre-determined size)
using the evolving psf on one image to match the other (i.e. min
(edge weighted?) pixel mse of differences), i.e. an iteration of the
formpsfnew= optimize(psflast, mse(psflast * I1, I2))