BFORE: Butterfly-Firefly Optimized Retinex Enhancement for Low-Light Image Quality Improvement
This academic paper introduces BFORE, a novel framework for enhancing low-light images by automatically tuning parameters in a multi-stage Retinex-based pipeline. Addressing the limitations of manual tuning in existing methods, BFORE employs a hybrid metaheuristic approach combining the Butterfly Optimization Algorithm (BOA) and Firefly Algorithm (FA). The process involves converting images to HSV color space, applying Adaptive Gamma Correction with Weighted Distribution, and performing adaptive denoising. BOA optimizes Multi-Scale Retinex parameters, while FA handles correction and denoising settings, using a switching strategy to balance exploration and exploitation. Evaluations on the LOL benchmark dataset show BFORE achieves a Peak Signal-to-Noise Ratio (PSNR) of 17.22 dB, outperforming traditional methods like Histogram Equalization and MSRCR. Notably, it surpasses the deep learning baseline RetinexNet in both PSNR and Structural Similarity Index (SSIM) without requiring training data. The study highlights a significant performance boost attributed to the hybrid optimization strategy. However, the submission history indicates that this version of the paper has been withdrawn by the author, Ahmed Cherif, suggesting potential revisions or retraction of the findings.
Wire timeline
BFORE: Butterfly-Firefly Optimized Retinex Enhancement for Low-Light Image Quality Improvement
This academic paper introduces BFORE, a novel framework for enhancing low-light images by automatically tuning parameters in a multi-stage Retinex-based pipeline. Addressing the limitations of manual tuning in existing methods, BFORE employs a hybrid metaheuristic approach combining the Butterfly Optimization Algorithm (BOA) and Firefly Algorithm (FA). The process involves converting images to HSV color space, applying Adaptive Gamma Correction with Weighted Distribution, and performing adaptive denoising. BOA optimizes Multi-Scale Retinex parameters, while FA handles correction and denoising settings, using a switching strategy to balance exploration and exploitation. Evaluations on the LOL benchmark dataset show BFORE achieves a Peak Signal-to-Noise Ratio (PSNR) of 17.22 dB, outperforming traditional methods like Histogram Equalization and MSRCR. Notably, it surpasses the deep learning baseline RetinexNet in both PSNR and Structural Similarity Index (SSIM) without requiring training data. The study highlights a significant performance boost attributed to the hybrid optimization strategy. However, the submission history indicates that this version of the paper has been withdrawn by the author, Ahmed Cherif, suggesting potential revisions or retraction of the findings.
cs.AI updates on arXiv.org