Machine Learning Guess – Object aware compression algorithms?

I got the idea from a machine learning lesson on the brain and neural networks.

I was wondering how a different part of the brain could process data from different sources when the path was altered.

One way would be to do image recognition on the the data and see if anything looks different from random.

Another way I guess is to compress the data with different compression algorithms.

This makes me wonder if you could set generate new compression algorithms by specializing the compression for the types of objects in the data.

So for music we would have a classical compression algorithm, a house or trance music compression algorithm.

Similarly for images one might have face image compression and nature photo compression and so on.

So my idea is simple.

Should there exist object aware compression algorithms?