Assignment 1 (Denoising Images)
What to do: You will apply 5 algorithms on the 4 images corrupted by different types of noises. Image01.jpg was corrupted by Poisson noise. Image02.jpg was corrupted by Gaussian noise. Image03.jpg was corrupted by speckle noise and Image04.jpg was corrupted by salt and pepper noise. Images are here https://drive.google.com/drive/folders/1JkKmcMw378NsyMfss-XHvcxyLzDNoBmX?usp=sharing
What are these 5 algorithms? You should choose (a) one linear filter, (b) two non-linear filters, and (c) two algorithms based on deep neural networks. You are not required to write any code for this assignment. You will use existing codes from the publicly available domain. Cite the sources (such as papers and/or software sites).
What to submit?
20 denoised images, name them as image1_alg1.jpg, image1_alg2.jpg, ..., image4_alg5.jpg.
Also, write a report with (1) the short descriptions of five algorithms you have chosen with citations of their sources and citations of codes you used and (2) noisy images and their denoised versions obtained with each algorithm and your observations about these results. Your report length should not exceed 5 pages formatted as single-spaced, single column, using 12pt font text. I encourage you to use latex. Overleaf is a great free service you can make use of.
Marking: I have the original images. I will numerically compare your submitted denoised images against the original images. I will use SSIM metric (https://en.wikipedia.org/wiki/Structural_similarity) for this purpose. I will use the average SSIM score in the formula: 60 + (your average SSIM for 20 images)*40 to compute your score for this part of the assignment This score will have 4% of the total 10% allocated for the assignment.
Another 6% will be devoted to your (maximum) 5-page report.
zip all files into a single file and name as firstname_lastname.zip and submit.