Fusion Deep Learning with Adaptive Gamma Correction (DLAGC) to Enhance Images in Low Light

Abstract

Object detection in dark areas or poor light images represents a big challenge in computer vision. The poor light images suffer from intensive noise, low contrast, and reduced visibility. Based on the affirmation, this paper proposes a fusion model based on deep learning with adaptive gamma correction (DLAGC).  It enhances the low-light image Based on the combination of the images that outcome from deep learning and adaptive gamma correction in pixel-level image fusion.  The deep learning estimated pixel-level adjustment curves of RGB channels.  Moreover, the adaptive gamma correction value is calculated based on the value of the image Luminance factor and the average color factor, resulting in a locally adaptive value with each pixel.  The proposed model DLAGC has demonstrated the ability to improve image quality by enhancing lighting, highlighting fine details, and reducing noise while maintaining natural color balance.  To evaluate the proposed model, two reference datasets (LOL and Brightening Train) and three non-reference datasets (DICM, LIME, UCID_V2).  The Experimental results show that the proposed model outperforms the state of the art of low light methods. The proposed model gets an average PSNR is 17.386, SSIM is 0.788, and FSIM is 0.92 for reference datasets. Meanwhile, the achieved average NIQE is 3.684 for nan-reference datasets. Therefore, the model provides a real-world solution for image enhancement under different lighting conditions.