Enhanced automated art curation using supervised modified CNN for art style classification

Area of Science:

• Computer Vision
• Artificial Intelligence
• Art History

Background:

• Traditional art classification relies on time-consuming and subjective human expertise.
• Existing automated methods lack the precision needed for nuanced art analysis.
• Need for objective and scalable solutions in art classification and curation.

Purpose of the Study:

• To develop and evaluate a supervised Modified Convolutional Neural Network (CNN) for automated art classification.
• To distinguish between major art styles and movements using visual features.
• To provide a more consistent and scalable alternative to manual art curation.

Main Methods:

• A Modified CNN model was designed to analyze features like color, texture, and composition.
• The model was trained on a custom dataset of 5000 artworks across five styles: Impressionism, Cubism, Realism, Abstract, and Surrealism.
• Performance was benchmarked against ResNet50 and VGG16, with feature visualization techniques (t-SNE, PCA, Grad-CAM) employed for analysis.

Main Results:

• The Modified CNN achieved an average classification accuracy of 93.0%.
• It outperformed ResNet50 and VGG16 in precision (93.5%), recall (92.8%), and F1-score (93.1%).
• Feature visualization confirmed style clustering and identified challenges in differentiating similar styles like Abstract and Surrealism.

Conclusions:

• Modified CNNs show significant potential for accurate and automated art classification.
• The model offers a scalable and consistent tool for applications in digital curation, education, and preservation.
• Future research will focus on dataset expansion, multimodal inputs, and enhanced explainable AI techniques.