Computational Aesthetics Based on Gaze Patterns

Gary R. Greenfield
Meeting Alhambra, ISAMA-BRIDGES Conference Proceedings (2003)
Pages 85–92


Generative art systems usually rely on the technique of user-guided evolution to enable "artists" to interactively search through populations of images in order to breed those images which show aesthetic promise. We consider algorithmic criteria for automating the aesthetic evaluation step. Our criteria are inspired by a recent description of techniques used for controlling the aesthetic reorganization of photorealistic imagery. Surprisingly, our approach seems to impose a rigid style on the images we evolve. A drawback to our method is that eventually it fails to clearly differentiate between non-degenerate and degenerate images. We consider how improvements might be made.