Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Eric Chen, Patricia Alves-Oliveira

TL;DR

This work introduces a patch-based spatial authorship attribution framework for human–robot collaborative paintings, addressing data-scarce environments and spatially varying authorship. It demonstrates that learning from 300×300 grayscale patches yields 88.8% patch-level accuracy and 86.7% painting-level accuracy across 15 paintings, outperforming texture-based and pretrained-feature baselines. The authors further leverage conditional entropy to reveal regions of mixed authorship in hybrid paintings, finding significantly higher uncertainty in collaborative regions, which reflects overlapping stylistic cues rather than misclassification. The methodology provides a sample-efficient foundation for forensic attribution in data-scarce human–AI creative workflows and points toward extensions to broader human–robot collaborations and process-level analysis.

Abstract

As agentic AI becomes increasingly involved in creative production, documenting authorship has become critical for artists, collectors, and legal contexts. We present a patch-based framework for spatial authorship attribution within human-robot collaborative painting practice, demonstrated through a forensic case study of one human artist and one robotic system across 15 abstract paintings. Using commodity flatbed scanners and leave-one-painting-out cross-validation, the approach achieves 88.8% patch-level accuracy (86.7% painting-level via majority vote), outperforming texture-based and pretrained-feature baselines (68.0%-84.7%). For collaborative artworks, where ground truth is inherently ambiguous, we use conditional Shannon entropy to quantify stylistic overlap; manually annotated hybrid regions exhibit 64% higher uncertainty than pure paintings (p=0.003), suggesting the model detects mixed authorship rather than classification failure. The trained model is specific to this human-robot pair but provides a methodological grounding for sample-efficient attribution in data-scarce human-AI creative workflows that, in the future, has the potential to extend authorship attribution to any human-robot collaborative painting.

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

TL;DR

This work introduces a patch-based spatial authorship attribution framework for human–robot collaborative paintings, addressing data-scarce environments and spatially varying authorship. It demonstrates that learning from 300×300 grayscale patches yields 88.8% patch-level accuracy and 86.7% painting-level accuracy across 15 paintings, outperforming texture-based and pretrained-feature baselines. The authors further leverage conditional entropy to reveal regions of mixed authorship in hybrid paintings, finding significantly higher uncertainty in collaborative regions, which reflects overlapping stylistic cues rather than misclassification. The methodology provides a sample-efficient foundation for forensic attribution in data-scarce human–AI creative workflows and points toward extensions to broader human–robot collaborations and process-level analysis.

Abstract

As agentic AI becomes increasingly involved in creative production, documenting authorship has become critical for artists, collectors, and legal contexts. We present a patch-based framework for spatial authorship attribution within human-robot collaborative painting practice, demonstrated through a forensic case study of one human artist and one robotic system across 15 abstract paintings. Using commodity flatbed scanners and leave-one-painting-out cross-validation, the approach achieves 88.8% patch-level accuracy (86.7% painting-level via majority vote), outperforming texture-based and pretrained-feature baselines (68.0%-84.7%). For collaborative artworks, where ground truth is inherently ambiguous, we use conditional Shannon entropy to quantify stylistic overlap; manually annotated hybrid regions exhibit 64% higher uncertainty than pure paintings (p=0.003), suggesting the model detects mixed authorship rather than classification failure. The trained model is specific to this human-robot pair but provides a methodological grounding for sample-efficient attribution in data-scarce human-AI creative workflows that, in the future, has the potential to extend authorship attribution to any human-robot collaborative painting.
Paper Structure (34 sections, 2 equations, 4 figures, 4 tables)

This paper contains 34 sections, 2 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Human Produced Art and Authentication
  4. Art Authentication and Authorship
  5. Computational Authentication
  6. Robotic and AI-Generated Art
  7. Robotic Art Systems
  8. Authentication of Robotic and AI-Generated Art
  9. Dataset Construction
  10. Paintings and Digitization
  11. Patch Extraction and Statistical Units
  12. Human Paintings
  13. Robot Paintings
  14. Hybrid Paintings
  15. Model Architecture
  16. ...and 19 more sections

Figures (4)

  • Figure 1: Example of the paintings from the dataset showing human-created paintings (a), robot-created paintings generated using the CoFRIDA framework schaldenbrand2024cofrida (b), and collaborative human-robot paintings (c).
  • Figure 2: The robotic painting system in a collaborative painting session, illustrating the workflow used to generate hybrid paintings.
  • Figure 3: Patch-level confusion matrices aggregated across all cross-validation folds. (a) Raw counts. (b) Per-class recall.
  • Figure 4: Entropy analysis comparing pure and hybrid paintings. (a) Violin plots show pure human and robot paintings exhibit similar low-entropy distributions (medians 0.11 and 0.13), while hybrid paintings display elevated uncertainty (median 0.18). (b) CDFs demonstrate systematic separation between pure and hybrid paintings. Dashed lines indicate median entropy for each category.