Super Resolution - Printable Version +- Gimp-Forum.net (https://www.gimp-forum.net) +-- Forum: GIMP (https://www.gimp-forum.net/Forum-GIMP) +--- Forum: General questions (https://www.gimp-forum.net/Forum-General-questions) +--- Thread: Super Resolution (/Thread-Super-Resolution)

Super Resolution - Occasionnel - 03-22-2021 Hello I am occasionnaly worker with GIMP and I am generally use it for photography. I will need to generate a very large image full of pixel and I discover a tuto for PS: Something like that with some actions I don't know in GIMP Bring all images into Photoshop as a stack of layers Resize the image to 200% width and 200% height using 'Nearest Neighbor' Auto align all the layers Average the layers by setting each layer's opacity to 1/layer number (the 1st layer will be 1/1 so 100% opacity, the 2nd layer will be 1/2 so 50% opacity, and the 4th layer will be 1/4 or 25% opacity, and so on). Sharpen the image using a Radius setting of 2, and a suitable Amount setting (we used 200% for the 4 image stack and 300% for the 20 image stack – the more images you stack the more amenable the composite will be to aggressive sharpening) Alternatively, for the fourth step you can convert all layers to a 'Smart Object' and change the stacking mode to 'Median'. This can help deal with ghosting from movement in your final image, but can also take longer to process. So looked for the same with linux tools (Gimp) but not able to find it out. Is the smart object or other options that not permit to do the same ? Thanks you all RE: Super Resolution - Ofnuts - 03-22-2021 These are really two different methods for two different things. The first method is mostly about reducing noise by averaging it out. No "smart layers" in Gimp but once you have the stack you can do a Layer ➤ New from visible and continue the processing with this. If there is a bit of difference between layers this could lead to improving the definition (IIRC this is done as "pixel shifting" in Sony high-end cameras) but using a plain average won't help much (in the Sony implementation they know exactly how they moved the sensor behind an immobile lens between the takes, but in you case it will be random moves with random geometry changes, so registering the layers to make them coincide will likely eat out the extra definition. The second method is mostly about removing transient "defects", typically (moving) people, when shooting something fixed (buildings, scenery). No median filter in Gimp itself but there is one in the GMIC filter suite (there is also an average filter).  If you feel adventurous there are also my own implementations of average and median in Python. RE: Super Resolution - Occasionnel - 03-23-2021 Thanks a lot for your comment !