This repo contains an implementation of an FFT-based compression scheme. This product takes in an image, and outputs a compressed representation with a user-defined amount of "lossiness." We take advantage of the symmetries of the resulting matrix of the Fourier transform of a two-dimensional data set, allowing us to throw away entire portions of the frequency domain of an image. We then recover these frequencies the process of decompression before taking the inverse Fourier Transform of the data to recover an approximation of the original image.
Clone this repo, then run python compression.py to generate 100 compressed images of lena.ppm. Filenames correspond to levels the percentage of the original frequencies contained in the image approximation. So lena01.ppm only contains 1% of the original frequency data (and she still looks great!).
View the images with a PPM viewer. ToyViewer for Mac OSX is an adequate product.