Automatic Representative View Selection of a 3D Cultural Relic Using Depth Variation Entropy and Depth Distribution...

Area of Science:

• Computer Graphics and Vision
• Digital Heritage and Cultural Informatics

Background:

• Selecting representative views of 3D cultural relics is vital for digital museum applications.
• Existing view-based methods struggle to balance human preference with effective feature description.
• There is no established definition for optimal single or multiple views in computer graphics.

Purpose of the Study:

• To develop novel, objective measures for selecting representative 3D cultural relic views.
• To improve the accuracy and efficiency of 3D shape retrieval and classification.
• To automate the process of generating acceptable display views for virtual heritage.

Main Methods:

• Defined two information entropy measures: depth variation entropy and depth distribution entropy.
• Generated canonical pose 3D models and captured depth maps using orthographic cameras.
• Calculated entropy measures on depth maps and selected views with maximum entropy.

Main Results:

• Identified optimal viewpoints based on information entropy, aligning with human perception.
• Developed a 3D relic representation using histograms of local descriptors from selected depth maps.
• Demonstrated high performance in 3D shape retrieval and classification with only four views.

Conclusions:

• The proposed entropy-based method effectively selects representative views for 3D cultural relics.
• The approach automates view selection, eliminating manual pose adjustment.
• The method shows strong retrieval and classification capabilities, even with a limited number of views.