This course is divided into two parts.
First, a methodological section focusing on two central tools for image restoration and editing: patch methods and generative neural networks. Patch methods are based on the premise that the local neighborhoods of digital images (patches) have self-similarity properties that can be exploited to improve image quality. Generative networks (auto-encoders, generative adversarial networks) enable image enhancement or synthesis, based on training on large databases.
In the second part, the course presents several applications from the field of computational photography, which involves overcoming the limitations of imaging devices by algorithmic means. The two methodological approaches seen initially will be implemented in this context.
The following applications will be detailed:
- High Dynamic Range (HDR) imaging
- Image deconvolution
- Inpainting