Abstract
Images generated by artificial intelligence models that operate in continuous color spaces can exhibit chromatic artifacts, such as color drift and edge feathering, when converted to a discrete, canonical color palette. A computerized post-processing method can address these issues. The method can involve dividing a source image into blocks using resolution-adaptive spatial partitioning, and a representative modal color can be determined for each block. A system can also identify a dominant background color of the source image and add it to a target palette to create an enhanced palette. A k-dimensional tree may then be used to map each block's modal color to the nearest color in the enhanced palette. This process can be used to convert images to a fixed palette, potentially reducing artifacts and decreasing the need for manual correction while improving the preservation of structural integrity.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Mhatre, Parag; Faggian, Nathan; Chaudhary, Kartik; Tendulkar, Ashish; Dhar, Gopala; Kang, Gyhun; and Shivhare, Neeraj, "Deterministic Palette Quantization Using Resolution-Adaptive Spatial Partitioning", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/10979