Tile a ton of these photos into one large image, grayscale it and threshold for high brightness, then apply a lossless image compression algorithm. Compare it w

Tile a ton of these photos into one large image, grayscale it and threshold for high brightness, then apply a lossless image compression algorithm. Compare it with the combined size of the same process on each image individually. I bet (excluding the amortized cost of file metadata overhead) the tiled image will be smaller in terms of bits per pixel, suggesting correlation. Also, the perplexity is 2^H(x) ~ 2^len(compressed(x)) when compressed optimally. You could compare it to an image of the same size where each pixel, on average, draws values from a uniform '2^H(x)'-sided die. So you could use this trick to compare it to other images, lets say.