IMAGE WARPING and MOSAICING

Mengyang Zhang

Recover Homographies

I used the following matrix to calculate the Homographies between two pictures. Since there may not be a perfect solution, so I used the least square solution

no images no images

Recover Homographies

I used the night street as an example to demonstrate the rectified images

no images
perspective from the left side
no images
original example photo
no images
perspective from the right side

I used another street corner

no images
perspective from the left side
no images
original example photo
no images
perspective from the right side

Blend

Here are three sets of merge

After wraping one picture onto another, I used another alpha channel to do the weighted blending so the transition are as smooth as possible, which works best if the alignment is on the same horizontal level or vertical level. Otherwise, the pictures will have some sharp edge on some degree

no images
Left Picture
no images
Merged Picture
no images
Right Picture
no images
Left Picture
no images
Merged Picture
no images
perspective from the right side
no images
Left Picture
no images
Merged Picture
no images
Right Picture

What I learned

The image rectification is pretty interesting and useful. It's hard to believe that we can the different sides of the building just from one picture