After solidifying the idea of trying to algorithmically compose photos of two distinct style of architecture, the idea arose that it would be great to preprocess the two styles and classify the images into their respective architecture style or epoch as a nice addition, given deadline time constraints.

This would add a preliminary stage to the composition algorithm, which would categorically classify a supplied bank of images of the two different styles in order to best match the images to compose.

Having zero knowledge of machine learning, it looks like one option is to write the neural network myself (and train it, assuming I can provide the needed amount of imagery) and failing that, train the cascade classifier in openCV.

Some research links or points of interest: