Computational Photography

Here are the results of my project 1.







I did the project using the SSD method and followed instructions.
However, some images didn't align even though the code was right.
I overcomed this by only matching the center portion of every picture and managed to get satisfactory results using that method.
Also, this method tremendously sped up the processing for high res photos.
Finally I had some auto cropping code that cuts out the ugly edges.

Here are the stationary jpg photos for my results. The page will be flooded with pictures.
Enter at your own risk.


On the side note... ...
I had a bug in my code and couldn't solve it for a really long time.
But with the help of my classmates and the TAs I managed to solve my problem.
Yay! Thanks guys. :D

One method of improving the images is definitely to remove the salt of pepper noise present. This can be easily done using the median filter as discussed in class.

To improve the contrast, we can process the image matrice and apply a linear equation on the numbers, such that numbers above 0.5 will be larger and numbers above 0.5 will be smaller.
For example: newNumber = factor*(currNumber - 0.5) + 0.5
This will cause the contrast to be bigger through linear interpolation.

To adjust the color levels, we can simply multiply the respective channels R,G,B by a constant factor before concatanating the 3 channels.

To achieve the best results, we might want to run a script that changes the factors and saves all images, allowing the artist to pick the best tweaked image among all the others.

What I did to remove the sides was to extract the middle values of the picture using matlab matrix indexing.