[Paper Review] Color: A Crucial Factor for Aesthetic Quality Assessment in a Subjective Dataset of Paintings

Seyed Ali Amirshahi, Gregor U. Hayn‐Leichsenring+2 more|arXiv (Cornell University)|Sep 19, 2016

Aesthetic Perception and Analysis7 references3 citations

TL;DR

This paper introduces a novel subjective dataset of 36 Western painters' works from the 15th to 20th century and investigates the role of color in aesthetic quality assessment. Using a classifier on color features related to perception, it achieves a 73% classification rate between color features and subjective aesthetic scores, demonstrating color's crucial role in computational aesthetics.

ABSTRACT

Computational aesthetics is an emerging field of research which has attracted different research groups in the last few years. In this field, one of the main approaches to evaluate the aesthetic quality of paintings and photographs is a feature-based approach. Among the different features proposed to reach this goal, color plays an import role. In this paper, we introduce a novel dataset that consists of paintings of Western provenance from 36 well-known painters from the 15th to the 20th century. As a first step and to assess this dataset, using a classifier, we investigate the correlation between the subjective scores and two widely used features that are related to color perception and in different aesthetic quality assessment approaches. Results show a classification rate of up to 73% between the color features and the subjective scores.

Motivation & Objective

• To develop a new, diverse dataset of Western paintings spanning multiple centuries for aesthetic quality assessment.
• To investigate whether color features correlate with subjective aesthetic judgments in a human-annotated dataset.
• To evaluate the predictive power of color-based features in classifying aesthetic quality using machine learning.
• To contribute to computational aesthetics by identifying key visual factors influencing human perception of beauty in art.

Proposed method

• The authors curated a dataset of 36 well-known Western painters' works from the 15th to 20th century.
• Subjective aesthetic scores were collected from human observers, forming the basis for evaluation.
• Two widely used color perception features—likely related to color distribution and harmony—were extracted from each painting.
• A supervised classifier was trained to predict subjective aesthetic scores based on these color features.
• Classification performance was evaluated using accuracy metrics on the predicted vs. actual aesthetic rankings.
• The study used a standard machine learning pipeline to assess the correlation between color features and human judgments.

Experimental results

Research questions

• RQ1 To what extent do color-based features predict human subjective aesthetic judgments in paintings?
• RQ2 How well can a classifier distinguish high- from low-aesthetic-quality paintings using only color features?
• RQ3 Are color features a more influential factor than other visual features in aesthetic quality assessment?
• RQ4 Does the correlation between color features and aesthetic scores hold across diverse artistic styles and periods?

Key findings

• The classifier achieved a classification accuracy of up to 73% when predicting subjective aesthetic scores based solely on color features.
• Color features showed a strong and significant correlation with human-annotated aesthetic judgments.
• The results confirm that color is a crucial factor in aesthetic quality assessment, even when other visual features are ignored.
• The dataset demonstrates consistent performance across diverse artistic periods and styles, supporting its validity for aesthetic research.
• The study provides empirical evidence that color perception features can effectively model human aesthetic preferences in paintings.
• The findings support the integration of color-based features as a core component in computational aesthetic models.