Viewport-Unaware Blind Omnidirectional Image Quality Assessment: A Flexible and Effective Paradigm

TL;DR

This work tackles the inefficiency of viewport-based OBQA for omnidirectional images by introducing a viewport-unaware paradigm, VU-BOIQA, that operates directly on ERP projections. It introduces three components—Adaptive Prior-equator Sampling ($APS$), Progressive Deformation-unaware Feature Fusion ($PDFF$), and Local-to-global Quality Aggregation ($LGQA$)—to extract patch-level information, robustly fuse multi-scale features under ERP distortions, and map local scores to a global quality estimate. Extensive experiments across four OIQA databases show competitive performance with significantly reduced complexity and demonstrate the method's adaptability to 2D-IQA. The work suggests that viewport-unaware models can offer practical, scalable IQA for both omnidirectional and planar images, with avenues for tighter integration of deformation metrics and a unified 2D/OIQA framework.

Abstract

Most of existing blind omnidirectional image quality assessment (BOIQA) models rely on viewport generation by modeling user viewing behavior or transforming omnidirectional images (OIs) into varying formats; however, these methods are either computationally expensive or less scalable. To solve these issues, in this paper, we present a flexible and effective paradigm, which is viewport-unaware and can be easily adapted to 2D plane image quality assessment (2D-IQA). Specifically, the proposed BOIQA model includes an adaptive prior-equator sampling module for extracting a patch sequence from the equirectangular projection (ERP) image in a resolution-agnostic manner, a progressive deformation-unaware feature fusion module which is able to capture patch-wise quality degradation in a deformation-immune way, and a local-to-global quality aggregation module to adaptively map local perception to global quality. Extensive experiments across four OIQA databases (including uniformly distorted OIs and non-uniformly distorted OIs) demonstrate that the proposed model achieves competitive performance with low complexity against other state-of-the-art models, and we also verify its adaptive capacity to 2D-IQA.

Figures (6)

• Figure 1: An intuitive comparison of the general OI processing procedure (left) and the proposed one (right). (left) Researchers usually extract viewports using a "computationally expensive" scan-path prediction method or relying on the "often unavailable" ground-truth viewing behavior; (right) We simplify this procedure substituted by extracting patches from the "visually deformed" OI directly.
• Figure 2: The framework of the proposed VU-BOIQA model, which mainly consists of three modules, i.e., an APS module for directly extracting patches from OIs in the ERP format, a PDFF module for jointly addressing the inborn irregular geometric distortion in ERP images and fusing multilevel features, and a LGQA module for mapping patch-wise quality to global quality.
• Figure 3: The diagram of the proposed APS module, which performs random sampling in both latitude and longitude directions based on the equatorial prior distribution.
• Figure 4: The deformation perception operation generates offsets for each sampling position through convolutional layers, dynamically adjusts the convolutional kernel’s sampling locations based on these offsets, and performs weighted aggregation of the sampled features.
• Figure 5: The architecture of the DAA module.