Advances in Image Demosaicing and Pansharpening

The field of image processing is witnessing significant advancements in image demosaicing and pansharpening. Researchers are exploring innovative approaches to improve the efficiency and accuracy of these techniques, particularly in resource-limited mobile devices and real-world applications. Notably, the development of lightweight neural networks and dynamic convolution strategies is enabling faster and more effective image processing. Furthermore, the integration of attention mechanisms and state space models is enhancing the representation of high-frequency features and improving the overall quality of reconstructed images.

Some noteworthy papers in this area include:

  • A lightweight Quad Bayer hybrid EVS demosaicing approach that achieves state-of-the-art performance while reducing parameter and computation costs.
  • A dynamic splitting convolution strategy for pansharpening that effectively extracts features and enhances network generalization.
  • A method for tackling cross-sensor degradation in pansharpening that offers improved generalization ability and low generalization cost.
  • A token-wise high-frequency augmentation transformer for hyperspectral pansharpening that preserves high-frequency components and reduces attention dispersion.

Sources

Lightweight Quad Bayer HybridEVS Demosaicing via State Space Augmented Cross-Attention

DSConv: Dynamic Splitting Convolution for Pansharpening

Training and Inference within 1 Second -- Tackle Cross-Sensor Degradation of Real-World Pansharpening with Efficient Residual Feature Tailoring

THAT: Token-wise High-frequency Augmentation Transformer for Hyperspectral Pansharpening

Built with on top of