# Binocular Rivalry and Fusion-Inspired Hierarchical Complementary Ensemble for No-Reference Stereoscopic Image Quality Assessment

## TL;DR

This paper introduces a new framework for assessing the quality of stereoscopic images by mimicking human binocular rivalry and fusion processes.

## Contribution

The novel Adaptive Selective Propagation strategy in a hierarchical Transformer enables dynamic binocular feature fusion inspired by human vision.

## Key findings

- The MSCE framework achieves state-of-the-art performance on four benchmark datasets.
- The ASP strategy effectively reinforces features based on binocular discrepancies using nonlinear gain control.
- The HCF module successfully integrates texture, structure, and semantic consistency in a unified quality-aware manifold.

## Abstract

No-reference stereoscopic image quality assessment (NR-SIQA) remains a fundamental challenge due to the complex biological mechanisms of binocular rivalry and fusion, particularly under asymmetric distortions. In this paper, we propose a novel framework termed Multi-Stage Complementary Ensemble (MSCE). The core innovation lies in the Adaptive Selective Propagation (ASP) strategy embedded within a hierarchical Transformer architecture to dynamically regulates the fusion of binocular features. Specifically, by simulating the human visual system’s transition from binocular rivalry to fusion, the ASP strategy applies nonlinear gain control to selectively reinforce features from the governing view based on binocular discrepancies. Furthermore, the proposed Hierarchical Complementary Fusion (HCF) module effectively captures and integrates low-level texture integrity, mid-level structural degradation, and high-level semantic consistency, leveraging ensemble learning principles, within a unified quality-aware manifold. Experimental results on four benchmark datasets demonstrate that the MSCE framework achieves state-of-the-art performance, particularly in terms of prediction consistency under complex asymmetric distortions.