During my research into Image Processing techniques, I came across a number of different approaches – with a vast majority being dead ends. I considered to make my report very technical, incorporating many ideas from signal processing, (more engineering perspective), such as Shannon’s Theorem or aliases. However, I found out that focusing on 3 ideas that I thought were simple was in-itself a report.
I focused on ensuring that I can develop an understanding of a few important ideas. Such as going into detail about the Discrete Fourier Transform, and how to use MATLAB’s formulation of it. I also made sure to visualize what MATLAB was doing, because all MATLAB gives is a simple formula. I needed to ‘see’ what MATLAB was doing – namely figuring out what the basis functions of the discrete Fourier transform looked like as images. This was difficult at first – and this told me that I should take a much closer look into the Discrete Fourier Transform, (DFT).
I felt that I understood the DFT fairly well, but I couldn’t justify signal processing techniques. I had to see which frequencies I was restricting and understand what impact that had on the image. The thing was, this isn’t too necessary to build a nice result, but it’s important to understand why and how it works. Especially if my project is on the DFT and its applications.
The key points of my research, and what I think I learnt the most from, was to ensure that I truly understood the use of the DFT on images. For example, the DFT gives you a way to decompose a signal into sinusoids of different frequencies. For images you can represent an image of n pixels by n images each with n pixels. That is interesting, but you can remove some of those n images to get a different result. Say, a blurrier image or a sharper one. But it is important to have a visual understanding of the removed images. Then if you see the final image, (minus the ones you removed), it shows dividend of your understanding. I liked this process, and it was the most interesting aspect of the project.
Sami Salem
The University of Adelaide
