How JPG Works

> This example shows a commonly used matrix of quantization factors , one for each basis image

Key point here that isn't spelled out in the text is that the low frequency stuff (top left) is weighted higher than high frequency stuff. After dividing by the quantization factora and rounding, most high frequency stuff rounds to zero (which is mentioned, it just wasn't spelled out). Higher quality levels don't have as large of factors there and preserve high frequency data.

(Edit: just got to where gets detailed later in the quality section, but still feel it should be mentioned where that matrix is first shown)