In this project, I wrote code to handle the projective transformation necessary to stitch images together, and combine them with the magic of (imperfect) alpha blending. It was refreshing to go back and re-visit some of those ole EE16A vibes in constructing the homography matrix by setting it up as a least-squares regression problem. This part I thought was genuinely quite cool, and gave me a much better intuition for understanding the different forms of image transforms. In combining the images, I thought deeply about what the relation between world coordinates and the coordinates of the transformed image pixels themselves.
Hiroshi Nagai, Tennis Camp | |
Wayne Thiebaud, San Francisco West Ridge | (Not a true rectification per se) |
It seems that the naive alpha blending still has a tendency to produces seams and other noticeable high frequency artifacts, unfortunately. Other issues are caused by auto white-balancing on the pictures from the lake (I only had my phone, but it was very picturesque), and what I believe to be vignetting from using older lenses.
Street scene | |
Lake | |
My room | |