This is the video of my classmate Oleg's hand. The video, though of poor quality, shows his pulse. Frequencies [0.5, 0.7].
Approach
I used the following approach to complete the assignment.
1) Created a Gaussian pyramid of the timeline of images by resizing the images by half subsequently.
2) Created a Laplacian pyramid by subtracting levels of Gaussian pyramid with upscaling by 2.
3) For each level of Laplacian pyramid, I filtered all the time vectors (which show how one a single pixel changes over time). I filtered vectors by applying the Fast Fourier transform on them, then multiplying by the filtering frequency vector. I amplified the frequencies I needed by 10 and then converted it back from the frequency domain, using Inverse fast Fourier transform. This was the result I needed.
4) Then, to rebuild the video, I recreated the image using pyramids, resizing and adding the filtered layers. The resulting images were written into a video file.
Bells and Whistles
None this time.
Difficulties
This project took an insane amount of time and effort for me. I faced a lot of diffuculties, and the more time I spent without getting any result, the more my frustration grew. However, I was really pleased when I finally completed it. I think I also received of good amount of understanding of how to work with image filters and pyramids. The main diffuculties were to properly build a Laplacian pyramid so that it wouldn't be almost completely black and to filter the time vector. Unfortunately, I didn't have time to work with all the RGB channels, so the final result is lower in quality than the original video. Â I would like to thank Professor Kitani and TA's De-An and Akanksha for helping me and answering my (sometimes not so smart) questions. Also thanks to my classmates from MSIT-SE who helped me with useful advice.