> Each 8×8 block is turned into a short DCT (Discrete Cosine Transform) feature vector; distance is measured there instead of raw pixels.

Was this your idea or the agent's? Because it feels peculiar to essentially describe JPEG, modulo the entropy coding, without mentioning its name. It was released in 1992 -- before C+C. Which makes me wonder why Westwood didn't just use JPEG too...

DCT's inventor: https://en.wikipedia.org/wiki/Nasir_Ahmed_(engineer)