Vision-Informed Flow Image Super-Resolution with Quaternion Spatial Modeling and Dynamic Flow Convolution

TL;DR

This work addresses flow image super-resolution (FISR) by bridging the domain gap between flow visuals and natural images. It introduces vision-informed FISR through quaternion spatial modeling to capture orthogonal UVW relations and dynamic flow convolution to exploit the droplet-like flow morphology. The proposed framework combines a Swin-Transformer–based flow feature extractor, quaternion layers, and morphology-aware sampling to deliver state-of-the-art HR flow reconstructions on DNS turbulence data, with comprehensive ablations validating the contributions. The approach significantly improves SR accuracy and turbulence perception, and the authors plan to release code and data to accelerate research on physics-informed flow reconstruction.

Abstract

Flow image super-resolution (FISR) aims at recovering high-resolution turbulent velocity fields from low-resolution flow images. Existing FISR methods mainly process the flow images in natural image patterns, while the critical and distinct flow visual properties are rarely considered. This negligence would cause the significant domain gap between flow and natural images to severely hamper the accurate perception of flow turbulence, thereby undermining super-resolution performance. To tackle this dilemma, we comprehensively consider the flow visual properties, including the unique flow imaging principle and morphological information, and propose the first flow visual property-informed FISR algorithm. Particularly, different from natural images that are constructed by independent RGB channels in the light field, flow images build on the orthogonal UVW velocities in the flow field. To empower the FISR network with an awareness of the flow imaging principle, we propose quaternion spatial modeling to model this orthogonal spatial relationship for improved FISR. Moreover, due to viscosity and surface tension characteristics, fluids often exhibit a droplet-like morphology in flow images. Inspired by this morphological property, we design the dynamic flow convolution to effectively mine the morphological information to enhance FISR. Extensive experiments on the newly acquired flow image datasets demonstrate the state-of-the-art performance of our method. Code and data will be made available.

Figures (3)

• Figure 1: Visual properties of flow images. Imaging Principle: the UVW velocities of the flow field that are set as RGB channels for imagery could be directly transferred into the orthogonal quaternion field. Morphological Property: The generalized droplet-like flow shape could be better analyzed by our designed dynamic flow convolution to provide the essential morphological information.
• Figure 2: The pipeline of our proposed vision-informed flow image super-resolution algorithm. The low-resolution flow images are first propagated into one convolution layer to generate the shallow feature. Then, through the sequential Flow Feature extraction Blocks (FFB), the vision-informed flow features are acquired. Subsequently, through the Quaternion Spatial Modeling (QSM) and a deep residual layer, the flow features are inputted into the Flow Image Reconstruction (FIR) module to obtain the final high-resolution flow images.
• Figure 3: The visual comparisons with advanced methods. HR: the high-resolution flow images. Ours: the results of our proposed method.