Viewpoint Assessment and Recommendation for Photographing Architectures.

Area of Science:

• Computer Vision
• Computer Graphics
• Computational Photography

Background:

• Assessing architectural photography viewpoint quality is challenging.
• Previous methods primarily relied on 2D image features, limiting accuracy.
• Online user annotations of landmark photographs provide valuable data.

Purpose of the Study:

• To develop a robust method for evaluating photographing viewpoint quality.
• To recommend optimal viewpoints for architectural photography.
• To leverage both 2D image and 3D geometric features for improved accuracy.

Main Methods:

• Collected internet photographs and 3D models of 15 world-famous architectures.
• Utilized image-model registration for viewpoint recovery.
• Employed a novel viewpoint clustering strategy to validate user preferences.
• Extracted 2D image and 3D geometric features.
• Developed a Support Vector Machine with 2-Kernel (SVM-2K) multi-view learner.

Main Results:

• Combining 2D and 3D features significantly improved viewpoint quality evaluation compared to using either alone.
• The proposed method achieved superior performance in viewpoint recommendation.
• Effectiveness demonstrated through extensive experiments.
• System validated for recommending viewpoints for rendering 3D architectural models.

Conclusions:

• Integrating 2D and 3D features offers a more reliable approach to viewpoint quality assessment.
• The developed system effectively recommends optimal viewpoints for architectural photography.
• The approach has practical applications in architectural design and visualization.