While the article is mostly good one glaring technical error annoys me. "The stronger the cosine transformation, the more compressed the final result" is simpl

While the article is mostly good one glaring technical error annoys me. "The stronger the cosine transformation, the more compressed the final result" is simply wrong. DCT (and the inverse DCT) are transforms that transform between "sample" and "frequency" domains, it's well defined should be perfectly reversible without compression (iirc the one in JPEG should be lossless given as many bits as the samples themselves). The trick of DCT based compression is that humans don't notice when information disappears from higher frequencies(also in _natural images_ there is often little data in high frequencies, often lots of 0 that can be immediately cut). So harder compression means removing more high frequency data from storage without it being too noticeable when reconstructing samples from the frequency domain at decompression. Conversly however, if you have "sharp edges" in the sample data you need more higher frequencies to reproduce the sharp edges without "ringing" artefacts (this is why you will see noisy blocks around text in highly compressed JPEG's with text since it runs out of bandwidth to adjust). The frequency domain values, and how compression affects removing various frequencies(black and white in the filter images) can be illustrated on the wikipedia filter comparsion example image below. (low frequencies are in the upper-left corner of the filter and spectrum images whilst higher frequencies horizontally are to the right and higher vertical frequencies are towards the bottom). https://en.wikipedia.org/wiki/File:DCT_filter_comparison.png https://en.wikipedia.org/wiki/File:DCT_filter_comparison.png https://en.wikipedia.org/wiki/Discrete_cosine_transform https://en.wikipedia.org/wiki/Discrete_cosine_transform (Mainly "Example of IDCT" section towards the bottom but also the preceding ones).