Tony Li's CS180/Project 1 Gallary

Methodology(Scroll Down to see pictures):

1. Determine the picutre's resolution is high or low by checking width pixel count. If picutre is high-res, downscale the picture twice by random-pooling.

2. Slice every pciture into 3 equal-size sub-pictures, where respectively would be filtered with B, G, R. Each sub-picture is also slightly longer than 1/3 of the original picture, so that they all contain the entire picutre object. 2.

3. Extract the central patch of every sub-picture. If the original picture is high-res, the central patch would be slightly bigger to capture the most information.

5.Define blanaced_metric for alignment: basically divide the patch into 4 x 4 or 6 x 6 subsections, and use SSIM on every subsections with corresponding subsection of another patch. The summation of SSIM scores would be the final score.

6. Each central-patch would be moved up/down/left/right to use blanaced_metric to see if it can fit into antother sub-picture's central patch.

7. Use the displacement vector (how much a central patch should move) to align the original filtered sub-picutres to each other.

8. Combine three aligned filtered sub-pcitures and return the colorized picutre.

Extra Credit: Skipped