MyHistory: Automatic Photo Album Creation

This paper proposes a method to create customized albums in six different categories, namely buildings, food, drinks, people, pets, and scenery and recommendations for users to remove near duplicate photos. The method comprises two main components, the recommendation system and the classification model. The recommendation system employs criteria such as blur, face, open-eye and smile detection to determine the best photo. The classification model is built with transfer learning by fine-tuning pre-trained networks, namely VGG19 and MobileNetV2. The results showed that both fine-tuned models achieved an accuracy of approximately 97% on the test dataset. This classification system is to help users create their personal albums automatically. The album creation process is streamlined through the classification model, which is then complemented by the recommendation system based on the photo’s category. The output of this method is a crafted album within the specified category, enriched by the removal of near-duplicate photos, offering users a personalized and refined album creation experience.