Related Experiment Video
Updated: May 31, 2025
04:23
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
1.7K
Enhancing Low-Light Images with Kolmogorov-Arnold Networks in Transformer Attention
Alexandru Brateanu1, Raul Balmez1, Ciprian Orhei2
1Department of Computer Science, University of Machester, Manchester M13 9PL, UK.
Sensors (Basel, Switzerland)
|January 25, 2025
Summary
This study introduces KAN-T, a novel Transformer network for low-light image enhancement (LLIE). KAN-T utilizes a unique attention mechanism to significantly improve visibility and detail in dark environments.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Deep Learning
Background:
- Low-light image enhancement (LLIE) is crucial for improving image sensor performance in poor lighting.
- Transformer models have shown significant promise in advancing LLIE techniques.
Purpose of the Study:
- To propose a novel Transformer network, KAN-T, for superior low-light image enhancement.
- To introduce an innovative attention mechanism inspired by the Kolmogorov-Arnold representation theorem.
Main Methods:
- Developed KAN-T, a Transformer network featuring a novel attention mechanism.
- Incorporated learnable non-linearity and multivariate function decomposition into the attention mechanism.
- Enhanced feature flexibility and contextual information capture.
Main Results:
- KAN-T achieved state-of-the-art performance in low-light image enhancement.
- Demonstrated superior quantitative and qualitative results compared to existing methods.
- Highlighted the effectiveness and broad applicability of the proposed approach.
Conclusions:
- The novel attention mechanism in KAN-T significantly boosts low-light image enhancement capabilities.
- KAN-T offers a powerful and versatile solution for improving visibility in challenging lighting conditions.
- The proposed method represents a significant advancement in the field of image processing.
