Turn abandoned patents
into your own filing
into your own filing
This invention describes a way to teach a computer system, specifically a neural network, how to make composite images look more natural. It works by showing the system an image that has been altered in a specific area, then having the system try to "harmonize" that altered image. The system learns by comparing its harmonized output to the original, unaltered image, with the goal of making the altered part blend in seamlessly. The training can also use a mask to focus the harmonization on a specific foreground region.
Why it matters: Filed before the widespread adoption of advanced generative AI models like diffusion models and improved GAN architectures. These newer techniques offer significantly better image synthesis and harmonization capabilities, making the training of such a system more effective and the resulting harmonized images of higher quality today.
AI gives you a few directions you could take this. Pick one, and we check whether your version is different enough to patent, then write the filing.
Reinvent this with AI
