Research & Papers

Patch-Based Spatial Authorship Attribution in Human-Robot Collaborative Paintings

A new patch-based framework uses commodity scanners to determine who painted what in collaborative artworks.

Deep Dive

Researchers Eric Chen and Patricia Alves-Oliveira developed a patch-based computer vision framework for spatial authorship attribution in human-robot collaborative paintings. Their method, tested on 15 abstract paintings, achieves 88.8% patch-level accuracy using flatbed scanners and cross-validation. It outperforms baseline methods by 4-20% and uses conditional Shannon entropy to quantify stylistic overlap in ambiguous hybrid regions, providing a forensic tool for documenting creative contributions in data-scarce human-AI workflows.

Why It Matters

Provides a forensic tool for artists and legal contexts to document contributions in the growing field of AI-assisted creative production.