Assignment 4:
Image Warping and Mosaicing
The goal of this assignment is to get your hands dirty in different aspects of image warping with a “cool” application -- image mosaicing. You will take two or more photographs and create an image mosaic by registering, projective warping, resampling, and compositing them. Along the way, you will learn how to compute homographies, and how to use them to warp images. There are five steps to this part of the project: Shoot and digitize images, Recover homographies, Warp the images, and blending the images into a mosaic.
Original images:
Rectified images:
Mosaic:
This following mosaics are blended using alpha blending. The mosaic of downtown Pittsburgh at night was interesting because the alpha blending couldn't compensate for the lighting difference between all three of the images in the center since all images overlapped. For comparison I've included the version without blending in the second image which just overlays the center image on top. Did alpha blending create the better picture? I also though the day/night blend of Pittsburgh was really cool and I was actually surprised it came out so clean considering the images were at significantly differing resolutions.
Created using autostitching of three images taken from Mt. Washington.
Created using manual point correlation and no alpha blending of the same three images.
Created using manual point correlation with a day and night image of Pittsburgh, both taken from Mt. Washington. This is neat because you can see the lights in the road lit up when they clearly are not lit up in the original image.
Created using autostitching of three images taken of plants.
Created using autostitching of the three images with the same person in the Fairfax lobby. This turned out much better than the manual approach.
Created using manual point correlation with three images of the same person taken in the Fairfax lobby. Apparently I'm really bad at selecting points manually.Autostitching: Interest Point Detection
Top 500 Strongest Interest Points |
Top 500 With ANMS |
|
|
Autostitching: Feature Descriptor Extraction
The following is 400 of the 500 feature descriptors that I extracted to be used in comparisons.
Autostitching: Outlier Detection
The following are images of points after outlier detection.(You might have to view the full size to see the points clearly, sorry!)
Autostitching: RANSAC
The points in blue are those that are kept after RANSAC is applied.(You might have to view the full size to see the points clearly, sorry!)
