Learning hybrid dynamic transformers for underwater image super-resolution

Abstract

Underwater image super-resolution is vital for enhancing the clarity and detail of underwater imagery, enabling improved analysis, navigation, and exploration in underwater environments where visual quality is typically degraded due to factors like water turbidity and light attenuation. In this paper, we propose an effective hybrid dynamic Transformer (called HDT-Net) for underwater image super-resolution, leveraging a collaborative exploration of both local and global information aggregation to help image restoration. Firstly, we introduce a dynamic local self-attention to adaptively capture important spatial details in degraded underwater images by employing dynamic weighting. Secondly, considering that visual transformers tend to introduce irrelevant information when modeling the global context, thereby interfering with the reconstruction of high-resolution images, we design a sparse non-local self-attention to more accurately compute self-similarity by setting a top-k threshold. Finally, we integrate these two self-attention mechanisms into the hybrid dynamic transformer module, constituting the primary feature extraction unit of our proposed method. Quantitative and qualitative analyses on benchmark datasets demonstrate that our approach achieves superior performance compared to previous CNN and Transformer models.

Frequency-Oriented Transformer for Remote Sensing Image Dehazing

“…Due to severe degradation, degraded images often contain severe blurring, and frequency domain information includes structure and edges, which is important for restoring clear images. However, most existing methods [ 4 , 6 , 13 , 14 , 15 , 16 ] explore spatial information to recover edges from degraded images, while frequency domain information is often overlooked in these methods.…”

“…Note that FPAE calculates the correlation between an element in the query and all elements in the key through a Hadamard product in the frequency domain rather than matrix multiplication in the spatial domain. Subsequently, to enhance the representation ability, we introduce prompt components [ 14 , 27 ] in FPAE to adaptively guide the model to focus on more important information. CRFN extracts information from different scales in features, integrates and processes global frequency domain information and local multi-scale spatial information in Fourier space, and reconstructs global content under the guidance of the amplitude spectrum.…”