It's pretty interesting how JPEG's can capture an image quite well even with low quality. From the slides, it seems to do this by surveying the image, and noticing where the low-frequency and high-frequency regions of the image are. High frequency areas usually can be altered much more due to the fact that humans are actually less susceptible to errors in high-frequency content, and the way these areas in the image are altered is by simply reducing the image data in those areas, effectively reducing our file size so that we don't need to use so much data for one image.

@Shonenmind I guess that high frequencies are really hard to pick up as they involve areas of high noise in an image, which is generally displeasing to look at.

Since jpeg is a block wise algorithm, it creates strange boundary between blocks when only filter out the low frequency.

I think the blockiness is (said to be) "typical" of jpegs (at least in the past, when bandwidth was more limited). It's harder for me to notice the color gradients (color squashing?) effects though on the non-luma channel