I had an idea I wanted to run past this community. Perlin noise uses a seed to generate noise, but what if we used the image of the noise to get the seed?

I thought it would be fun to see if a grouping of pixels or a photograph of some kind would have a seed in perlin noise. How would we do that though? We could use a brute force method, but it would be slow and consume a lot of power. a reverse engineering approach could work but would be quote technical and very messy. The brute force method would require the seed to have bounds. The reverse engineering would probably be problematic with local minima and maxima. What method would be better? I thought it would be a good problem to tackle