I thought it would be interesting to see how transmitting only the 251 (byte-size) highest-energy coefficients in the entire image would fare, which is somewhat

I thought it would be interesting to see how transmitting only the 251 (byte-size) highest-energy coefficients in the entire image would fare, which is somewhat of an idealization of the method used by the codecs in this comparison, but the result is not visually as good as either: http://i.imgur.com/TJaMs0f.png http://i.imgur.com/TJaMs0f.png This is a cheap imitation of an encoder: there is no entropy coding, but neither is there consideration for scan order, so there are some biases for and against. Assuming an entropy coder could reduce the bits to 67% there's a little improvement: http://i.imgur.com/ILY23uM.png http://i.imgur.com/ILY23uM.png Comparing them, it seems the block partitioning and filtering employed by DLI and H.265 become much more perceptually important than the raw mathematical effectiveness of the transform under bit starved rates like this.