Viewpoint strategies and quality evaluation for 3D reconstruction of exhibition spaces

Digital exhibitions require clear representations of spatial environments and exhibits, and optimized viewpoint selection can help improve reconstruction quality. Using the Library scene as an example, this study compares three viewpoint selection strategies—uniform sequence sampling, farthest-point sampling from the first viewpoint, and farthest-point sampling from a centroid-based starting viewpoint—by reconstructing the scene under identical configurations and evaluating 115 views. The mean Peak Signal-to-Noise Ratio (PSNR) values for the three groups are 22.7395, 22.7906, and 22.9666 dB, respectively, while the corresponding Structural Similarity Index (SSIM) values are 0.72130, 0.72703, and 0.72707. The centroid-based starting group achieves the highest pixel-level consistency, while its structural similarity is close to that of the first-viewpoint group. Accordingly, the centroid-based starting strategy is preferred when prioritizing average image quality.