Abstract
Redacting objects from high-resolution, wide-aspect-ratio images, such as panoramas, can be challenging, as certain automated methods may introduce resolution degradation, unpredictable cropping, or visual artifacts. Systems and methods can utilize an iterative pipeline that combines object detection with constrained generative inpainting and geometric alignment. A process may, for example, identify and mask target objects, provide a version of the image with blurred masked areas to a generative model for inpainting, and then geometrically align the generative model's output to correct for distortions. The aligned output can be used as a patch, selectively compositing the generated content into the masked regions of the original image. This selective patching approach can help maintain the original image's dimensions, resolution, and fidelity by primarily modifying the areas that require redaction.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Leung, Mira, "Object Redaction for High-Resolution Images Using Geometrically Aligned Generative Patches", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/10694