Adaptive Hypergraph Convolutional Network for No-Reference 360-degree Image Quality Assessment

TL;DR

This paper introduces an adaptive hypergraph convolutional network that models hierarchical features and complex interactions between viewports for improved no-reference 360-degree image quality assessment.

Contribution

It proposes a novel hypergraph-based model that captures high-order viewport interactions and incorporates content characteristics, addressing limitations of existing GCN methods.

Findings

Outperforms state-of-the-art models on public 360IQA datasets.

Effectively models high-order interactions between viewports.

Utilizes hierarchical features aligned with human visual perception.

Abstract

In no-reference 360-degree image quality assessment (NR 360IQA), graph convolutional networks (GCNs), which model interactions between viewports through graphs, have achieved impressive performance. However, prevailing GCN-based NR 360IQA methods suffer from three main limitations. First, they only use high-level features of the distorted image to regress the quality score, while the human visual system (HVS) scores the image based on hierarchical features. Second, they simplify complex high-order interactions between viewports in a pairwise fashion through graphs. Third, in the graph construction, they only consider spatial locations of viewports, ignoring its content characteristics. Accordingly, to address these issues, we propose an adaptive hypergraph convolutional network for NR 360IQA, denoted as AHGCN. Specifically, we first design a multi-level viewport descriptor for extracting hierarchical representations from viewports. Then, we model interactions between viewports through hypergraphs, where each hyperedge connects two or more viewports. In the hypergraph construction, we build a location-based hyperedge and a content-based hyperedge for each viewport. Experimental results on two public 360IQA databases demonstrate that our proposed approach has a clear advantage over state-of-the-art full-reference and no-reference IQA models.