Line Integral Convolution

It's not done with mirrors, but with Line Integral Convolution: you filter an image by averaging pixels along a curve segment that follows a given vector field. The technique was first proposed by Cabral and Leedom in SIGGRAPH '93 (pp. 263-270).

If you're in a hurry, you can just approximate the curve with a straight line, as I've done here. The top image was produced using low-frequency Perlin noise for the vector field. The bottom image was prefiltered by adding colored noise, and then filtered with LIC using a swirling vector field with fractal Perlin noise added for variation.

If you like seeing this kind of stuff done to people's faces, you should check out the work of a genius: Ralph Steadman's Paranoids. It's a book of celebrity portraits that he makes by smooshing around the guts of Polaroid photos as they develop.