From a practical standpoint it is computationally complex to make a meme filter. You basically have to make a filter for every single meme maker, and that filter also has to work in real time. The computational complexity of filtering memes is ridiculous. Custom FPGA's, the system has to be distributed to all datacenters, it's gets expensive fast.

A Practical Guide to Image Filters

The Dirty Ways

Flag a user as posting 'harmful content' and filter every image they post (shadowban)

The Lazy Ways.

Save the image, check all images against the image and filter the image if there is a match.

Save the metadata (filename, size, hash), check all images against the image and filter the image if there is a match.

The more interesting ways.

Use Feature Recognition (expensive, complex, difficult to rollout and requires significant technical expertise)/OCR

To beat the lazy ways, change a pixel, change the metadata and give it another go.

To beat the more interesting ways, you just have to keep changing what you're doing and make it costly to actively filter you.