Presently, drawing compares the images at a resolution of 64x64, by pixel and by color. The image is first scanned for non-alpha pixels and takes note of how many pixels of each color are present, and awards a bonus or penalty based on whether they match. Matches are weighed more generously than misses.
Then, the colors that were found are iterated through, and the amount of ink used per color is compared. Overshooting and undershooting by significant margins induce penalties. (Although the system is currently a bit flawed in that the preview image strokes are wider after scaling, so slightly overdrawing is benefitial as I had forgotten to properly scale this part of the calculation. But I digress.)
The two steps are then combined. Currently the combination favors ink usage over color matches (step 2 over step 1), working well for earnest attempts but doing little to combat single-line submissions from netting a low sum of income, or a vague scribble of appropriate ink amounts netting a hefty sum in comparison.
At the end, benefits based on the player's popularity and customer's inherent biases modify the result.