First please make sure you have read the supervision notes on expectations of learning at different levels
Perlin Noise
Perlin noise is the goto approach for undergrads (and more generally) when developers want to build a nice large landscape algorithmically. Its predicatable, well understood and has been extensivly written about.
- https://en.wikipedia.org/wiki/Perlin_noise
Its due to this popularity that I really reccommend that you dont use this as the central technology of your dissertation. Complete tutorials exist online showing you how to build realistic-ish landscapes with very little effort in modern game engines. Indeed Unity has a built in command to generate it.
So, to demonstrate sufficient learning at dissertation level you need to move your work far beyond this. Perhaps noise like this exists within your submission, its a great tool for naturalistic seeming textures, shapes and movement, but implementing it shouldnt be the main focus of your work (unless you have something really special in mind).
Model Synthesis (also Wave Function Collapse)
Model Synthesis more popularly known as Wave Function Collapse is is a popular approach to generating large scale self similar landscapes from minimal data. Indeed there are so many materials on it online that the challenge for a Masters Student is to find a way to demonstrate the apropriate level of learning. If you are considering a dissertation making use of this at MSc level then you should degfinately either have it as a minor tool in a greater system or otherwise consider a way to innovate or extend in this area.
Original Work
- https://paulmerrell.org/research/
- https://paulmerrell.org/wp-content/uploads/2021/06/model_synthesis.pdf
- https://paulmerrell.org/wp-content/uploads/2021/06/thesis.pdf
Easier to Grok
- https://en.wikipedia.org/wiki/Model_synthesis
- https://github.com/mxgmn/WaveFunctionCollapse
Successive Innovations and Applications
- Using Wave Function Collapse and Other Algorithms
- https://arxiv.org/abs/2308.07307
- https://www.iccs-meeting.org/archive/iccs2025/papers/159090105.pdf
- https://ouci.dntb.gov.ua/en/works/4zob5GKl/
- For Navigation Rather than Generation
- https://www.jsr.org/index.php/path/article/view/1626
- WFC Over Graphs rather than Grids
- https://www.jstage.jst.go.jp/article/transinf/E103.D/8/E103.D_2019EDP7295/_pdf
- https://ieeexplore.ieee.org/document/8848019
Generative Adversarial Neural Networks
Basically Deep Convoluted Generative Adversarial Networks can generate landscapes to match a sample set.
- https://github.com/liquidnode/neural_terrain_2
- https://cvanbattum.com/projects/heightmap-gen
- https://hal.science/hal-04266751/document
- https://eprints.whiterose.ac.uk/id/eprint/153088/1/Real_world_Textured_terrain_generation_using_GANs_1_.pdf
- https://arxiv.org/pdf/2403.08782v1
Conferences on PGC
https://pcgworkshop.com/database.php