Unleashing Attributes-content Adaptation with Multi-color Spaces for Food Photo Aesthetic Assessment

The rise of social media and digital food culture has elevated the importance of food photography, making aesthetic assessment a key area of interest. This study introduces a novel framework for evaluating the aesthetic quality of food photos by integrating attribute-content adaptation with multi-color space feature representations. Addressing the need for more accurate and nuanced aesthetic assessment models, our method combines high-level attributes with detailed content features. We leverage convolutional neural networks to extract these features, which are then fused with color features and enhanced by a proposed classifier network. Our experimental design includes comprehensive comparisons against baseline models that only use attribute-content adaptations and various classifiers, such as the support vector machine and random forest. Results on public datasets demonstrate that our proposed framework consistently outperforms baseline models, achieving the highest accuracy and F1-score metrics. An ablation study further reveals that the fusion of multiple color spaces significantly enhances performance compared with individual color spaces or no color space integration, underscoring the critical role of color diversity in aesthetic evaluation.