A homotopy-CNN framework for continuous deformation in low-light image enhancement.
Saved in:
| Title: | A homotopy-CNN framework for continuous deformation in low-light image enhancement. |
|---|---|
| Authors: | Shivam Kumar Jha, S. (AUTHOR), Mohana, N. (AUTHOR) |
| Source: | Connection Science. Dec 2025, Vol. 37 Issue 1, p1-11. 11p. |
| Subjects: | Convolutional neural networks, Image enhancement (Imaging systems), Image processing, Image quality analysis, Homotopy theory, Contrast effect |
| Abstract: | In this paper, we propose a homotopy-guided image enhancement framework for Low-Light Image Enhancement (LLIE), where a low-light image is progressively transformed into a well-lit version using a parametric deformation model. The enhancement pipeline incorporates a convolutional neural network that predicts structure-aware filtering parameters, which are applied via a 12-directional convolution kernel fusion, followed by adaptive gamma, contrast, and saturation refinements. Instead of relying solely on reference-free or black-box learning, the model leverages perceptual quality metrics such as SSIM and PSNR during training to optimise enhancement along the homotopy path. Notably, we explore the modulation of brightness across the range $ b \in [-2.0, 2.0] $ b ∈ [ − 2.0 , 2.0 ] , observing that $ b \leq 0 $ b ≤ 0 yields outputs closer to ground-truth references, while $ b \geq 0 $ b ≥ 0 enhances perceptual vividness. Experiments on the LOL dataset show that the method produces superior visual quality and strong quantitative performance, all while remaining efficient on standard CPUs without the need for GPU acceleration. [ABSTRACT FROM AUTHOR] |
| Copyright of Connection Science is the property of Taylor & Francis Ltd and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.) | |
| Database: | Psychology and Behavioral Sciences Collection |
|
Full text is not displayed to guests.
Login for full access.
|
|
| Abstract: | In this paper, we propose a homotopy-guided image enhancement framework for Low-Light Image Enhancement (LLIE), where a low-light image is progressively transformed into a well-lit version using a parametric deformation model. The enhancement pipeline incorporates a convolutional neural network that predicts structure-aware filtering parameters, which are applied via a 12-directional convolution kernel fusion, followed by adaptive gamma, contrast, and saturation refinements. Instead of relying solely on reference-free or black-box learning, the model leverages perceptual quality metrics such as SSIM and PSNR during training to optimise enhancement along the homotopy path. Notably, we explore the modulation of brightness across the range $ b \in [-2.0, 2.0] $ b ∈ [ − 2.0 , 2.0 ] , observing that $ b \leq 0 $ b ≤ 0 yields outputs closer to ground-truth references, while $ b \geq 0 $ b ≥ 0 enhances perceptual vividness. Experiments on the LOL dataset show that the method produces superior visual quality and strong quantitative performance, all while remaining efficient on standard CPUs without the need for GPU acceleration. [ABSTRACT FROM AUTHOR] |
|---|---|
| ISSN: | 09540091 |
| DOI: | 10.1080/09540091.2025.2546911 |
