GeoRank360-CNN: Spherical Geometry-Aware Visual Quality Assessment and Optimization for 360° Immersive Digital Media Scenes
International Journal of Advanced Computer Science and Applications · 2026 · 17 (10)
DOI: 10.14569/ijacsa.2026.0171084 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Advanced Computer Science and Applications, 2026. Verfasst von Weiya Yang und 1 weiteren Autor:innen.
Abstract (Originalsprache)
In 360° immersive digital media, visual quality assessment is crucial for maintaining perceptual fidelity, as compression distortion and equidistant cylindrical projection jointly affect the viewing quality. This study proposes a spherical geometric perception framework named GeoRank360-CNN for all-round image quality assessment and quality-guided optimization. This framework uses ConvNeXt-Tiny as the main spatial backbone network and EfficientNetV2-S as an auxiliary semantic information provider. The quality prior fusion branch further captures compression-sensitive cues, including high-frequency information, gradients, local contrast, saturation, and ERP seam discontinuity. To consider the non-uniform geometric characteristics of equidistant cylindrical projection images, this study introduces a spherical geometric perception convolution module, which combines anisotropic convolution branches, latitude condition gating, and spherical area perception pooling. The resulting features are used to estimate bounded quality correction and are combined with strong quality anchor boxes for final MOS prediction. This study introduces spherical rotation consistency learning and content relative quality ordering as auxiliary training constraints. Experiments on the CVIQ database use content-disjoint segmentation to avoid scene leakage. The PLCC of GeoRank360-CNN is 0.421, the SRCC is 0.377, the KRCC is 0.250, and the paired ranking accuracy is 0.890. The results of ablation experiments confirm the complementary effects of quality priors, spherical modeling, and sorting constraints. Quality-guided refinement maintains or improves PSNR and SSIM while providing conservative local enhancement.
Zitieren
APA
Yang, W., & Chen, L. (2026). GeoRank360-CNN: Spherical Geometry-Aware Visual Quality Assessment and Optimization for 360° Immersive Digital Media Scenes. International Journal of Advanced Computer Science and Applications, 17(10). https://doi.org/10.14569/ijacsa.2026.0171084
Vancouver
Yang W, Chen L. GeoRank360-CNN: Spherical Geometry-Aware Visual Quality Assessment and Optimization for 360° Immersive Digital Media Scenes. International Journal of Advanced Computer Science and Applications. 2026;17(10). doi:10.14569/ijacsa.2026.0171084
BibTeX
@article{Yang2026,
author = {Yang, Weiya and Chen, Liang},
title = {GeoRank360-CNN: Spherical Geometry-Aware Visual Quality Assessment and Optimization for 360° Immersive Digital Media Scenes},
journal = {International Journal of Advanced Computer Science and Applications},
year = {2026},
volume = {17},
number = {10},
doi = {10.14569/ijacsa.2026.0171084}
}
EUVIMED