Project 5: IMAGE WARPING and MOSAICING
Shide Li
Part A: IMAGE WARPING and MOSAICING
Shoot the Pictures
I took two sets of photos, each from a same point of view but different directions, with overlapping areas to define corresponding points.
1-1
|
1-2
|
2-1
|
2-2
|
Recover Homographies and Warp the Images
For each pair of images, I computed the corresponding H matrix to warp the second image into the first. Then I warped the images using the H matrix. Here are the results:
1-warped
|
2-warped
|
Image Rectification
Using the warping technique above, we can rectify images to desired perspectives. For the example below, I warped the airpods case into a square, so it shows the perspective from straight above.
Original
|
Rectified
|
Blend the Images into a Mosaic
For each pair of images, I used the corresponding H matrix and blended the second image into the first, using a wrighted average (p = 0.5) for the overlapping areas. The overlapping areas still appear to be blurry, which might be because of too few points of correspondences for least square. I will try to improve the results after the checkpoint.
1-Mosaic
|
2-Mosaic
|
Conclusion
The coolest thing I learned from this part is how to compute transformations to recover homographies. It is interesting to warp images into different perspectives from the actual angle. The most important part I would say is to devide the x and y coordinates by the w for each point. It is very easy to omit and would make results way off.