CoE 3SK3 Course Project. Image Interpolation. Phase II. Due date: April 8, 2013.
Phase II of the course project is for you to explore further on the topic of image ...
CoE 3SK3 Course Project Image Interpolation Phase II Due date: April 8, 2013. Phase II of the course project is for you to explore further on the topic of image interpolation (also called super-resolution in research literature), improve the results of bicubic interpolation that you have obtained in Phase I, and extend the scope of applications. You are required to perform the following tasks: 1. Compare the bicubic interpolation algorithm, the bilinear interpolation algorithm, and the Soft-decision Adaptive Interpolation (SAI) algorithm, which can be downloaded from Prof. Wu’s website, in both objective quality (using the SNR metric) and subjective quality (perceptual quality).
In order to compute the SNR values of interpolated images in the performance evaluation, you need to simulate the low-resolution image by down-sampling an original image (the true 2D signal), so that you know the ground truth when computing the SNR values of interpolated images produced by different algorithms.
Your evaluation needs to be thorough and have statistical significance. Therefore, you need to collect a diverse sample set of at least 10 images using flickr and/or google image search. The sample set should include images of various degrees of details, from simple (e.g., computer generated logo image or the alike) to complex (e.g., close-up shots of intricate objects, see http://www.flickr.com/photos/26907150@N08/with/8033281240/#photo_8033281240).
Run different algorithms of the comparison group on the above test images, and tabulate SNR values of the interpolated images by different algorithms. Draw conclusions on and document your findings: rank the tested algorithms in average SNR and visual quality; comment on how the evaluated algorithms behave on different image structures, such as smooth shades, sharp edges, fine textures, and etc.
Summarize the advantages and disadvantages of the tested algorithms; contrast the tested algorithms in design objectives and explain why they perform differently in SNR and visual quality.
2. Generalize your work in task 1 from grey scale images to color images. 3. Generalize your work in tasks 1 and 2 from images to videos (optional).
Deliverable: A written report to document the design of your experiments, the choice of your test images, your empirical observations, and present your analysis and conclusions as required above. NOTE: To make points clear in your explanations or/and arguments, use output images of tested algorithms, letting empirical evidence speak for itself.