The idea of superresolution is to combine multiple images of low resolution to one image of high resolution. This means, using superresolution images with higher resolution than the native resolution of the camera can be created. Also image noise will be reduced in the process.
1 | Record a series of 10 to 30 images of something. |
|---|---|
2 | Sort out blurry or shaky images |
3 | Upsample every image by dividing each original pixel into four pixels or by interpolating the pixels. |
4 | Find the feature pattern in every image. This can be done by a brightness or hue threshold. More advanced procedures will increase the result quality. |
5 | Determine the exact position of the pattern in all images. |
6 | Align all the images so the patterns are all in the exact same location. For a start, basic shifting of the x and y coordinates will do the job. To increase the result quality, techniques like rotating in multiple axes can be applied. |
7 | Calculate the final pixel colors of the output image as the averagerage of the pixel colors of the aligned images. |
The leftmost image is one of 13 images taken of the moon with a 5MP smartphone camera without additional optics. It has been upsampled already.
The second image is the output of the superresolution algorithm. Notice the reduction in noise and aliasing distortions.
The third image is the second image edited in Gimp 2.
The fourth image is an image taken by a telescope for comparison.
