I took pictures from my Iphone. One pair was taken during the day in Wheeler Hall, the other two pairs were taken from the roof of an Apartment at Berkeley.
|
|
|
|
|
|
The equation to solve for the eight parameters of the homography can be derived from the definition of homography projection
From this we can formulate the parameters as the variables to solve for in a linear equation. h_33 in this equation can be set to 1, with hypothetical, appropriate scaling of the other variables. The with their correspondences are presented below.
 |
 |
 |
 |
 |
 |
Per the suggestion in the homework instructions, I took a rectangular section of the image that did not have a rectangular 2D projection, and I made it properly rectangular in the 2D projection. To apply the proper shift, I anchored one of the correspondences, and to scale the edges, I used the distances of neighbor correspondences as the rectangle edges. I reason that the scaling and shift don't have to be too accurate, because the image will simply be shifted or scaled some from optimal.
 |
 |
| original image w/square | rectified image |
 |
 |
To construct the mosaic, one image in the pair was warped to the perspective of the other image. The warp needed three stages: 1) The proper corners of the original image was warped to the perspective of the second image, in order to establish the warp boundaries. 2) The pixels inside the boundaries were inverse warped to the original image pixel. 3) The image is rebounded standard image indexing
 |
 |
 |
I tried laplacian blending as well as averaging of the intersection. Surprisingly, laplacian blending aggrevated the edges.
 |
 |
| laplacian | simple average |
 |
 |
 |
 |