That does look promising. My maps are made by random walks -- three traces of two-dimensional fractional Brownian motion with different fractal dimensions and independent random skew. Your Perlin map looks like it could have been done with my generator; probably the gradient is providing skew, and the Perlin noise is effectively being summed over successive steps to produce Brownian motion -- the same general effect, but from a much more accessible procedure

Try compositing multiple walks with different gradients (direction and magnitude) and different noise fields, and you'll probably get results equivalent to mine.

Nicely done!