University coursework. This was the smarter version of the convolution coursework. It used the premise that corners are what make images recognisable. Straight lines bring little value. Flat surfaces bring little value.
The high level explanation of what SIFT does is difficult to provide in a few words because it actually is more of a pipeline. Nonetheless, the most revolutionary feature about it is the use of blur filters. More specifically, blurring the image to analyse by a little bit, then doing it again but blurring a bit more. You get 2 blurred images, one more blurry than the other. Turns out, if you do a pixel-by-pixel subtraction of those two slightly different images, what you get is corners and edges *mind blown*
Run the program on any image and the program will show you locations and orientations of relevant corners
This was the coolest project out of my last year of uni. I liked it because this is actually relevant right now with self driving cars and general image recognition. As in everything, it is nice when you can connect theory with real world problems.