POI Summarization by Aesthetics Evaluation From Crowd Source Social Media

Area of Science:

• Computer Science
• Image Processing
• Computer Vision

Background:

• Place-of-Interest (POI) summarization is crucial for image retrieval.
• Existing methods often lack consideration for aesthetic quality and camera viewpoint diversity.

Purpose of the Study:

• To propose a system for POI image summarization that integrates aesthetic evaluation and camera distribution diversity.
• To enhance the user experience in image retrieval by recommending visually appealing and diverse POI images.

Main Methods:

• Developed a coarse-to-fine POI clustering approach to generate visual albums.
• Created 3D models from social media images for each album.
• Implemented a crowd-sourced saliency model for candidate photo selection and aesthetic measurement.
• Utilized 3D model saliency to generate saliency maps and an adaptive image adoption strategy.
• Combined aesthetic scores and viewpoint diversity for final image recommendations.

Main Results:

• The proposed system successfully generates visual albums and 3D models.
• Candidate photos were selected based on a crowd-sourced saliency model.
• Aesthetic measurement was improved using crowd-sourced saliency detection on 3D models.
• The system effectively recommends images balancing aesthetics and diverse camera distributions.

Conclusions:

• The developed POI summarization system effectively addresses both aesthetic quality and viewpoint diversity.
• This approach offers a significant improvement in image retrieval for POIs.
• The integration of 3D modeling and crowd-sourced saliency enhances image selection accuracy and relevance.