Project: Cellular Automata Image Style Transfer
Goal of the project:
To take an input image and build a probabilistic cellular automaton that is most likely to generate that image.
Summary of work:
I first implemented the ConvChain image synthesis algorithm in the Wolfram Language. After that, I expanded it to allow synthesis from any linear combination of multiple samples. Then I further expanded the algorithm to perform CA-powered style transfer, mixing the style of an image with the content of an other.
Results and future work:
With a successful implementation of the algorithm for single- and multiple-sample synthesis and style transfer, the next steps would mainly be performance upgrades. Further work could be done to reduce the probabilistic nature of the algorithm by having it produce initial conditions and rules of the CA that can generate the given image.