Adaptive Machine Learning Framework for Personalized Photograph Quality Evaluation
Nandagopal Parise · 2025
This paper introduces a machine learning-driven framework for evaluating photograph quality, catering to both developers and end-users. By integrating low-level algorithmic features with high-level subjective qualities, the framework enables personalized photograph assessments. Developers can seamlessly incorporate new features, while users can adjust the relative importance of high-level attributes like exposure or sharpness. The system employs Support Vector Regression (SVR) to link low-level features with high-level attributes and utilizes Linear Regression for overall quality estimation. Data from Amazon Mechanical Turk provides ground truth for training. Implemented in a practical application, this approach outperforms untrained models, offering improved accuracy and adaptability.