An Enhancemnt of Low-Light Images Using Hierarchical Pyramid Attentive Network and Hierarchical Feature Fusion

The development of the low-light images assists efficiently in high visioned tasks which often fails in the poor illumination conditions. Most of the old methods, applied has enhanced the low-lighted degrade images which resulted in the undesirable enhancement including the intensive noise, blurred details and slanted color. This paper is used for finding alignment to an end-to-end image along with enhancement of the perceptual information by pixel wise. To reduce the differences in style, Hierarchical Pyramid Attentive Network (HPANet) was used, which results in the betterment of the illuminations and the reveal exposed information’s. After aligned images, a hierarchy pyramid enhancements subnetwork is used for optimizing the quality of the image, which helps for removing the noises, results in enhancement low light images. A Hierarchical Feature Fusion (HFF) module is proposed which results in the high resolution of the images in perceptual qualities. Extensive experiments have also performed that effectiveness of HPANet-HFF methods on various datasets have outdone the existing methods in reproduction of the color. According to the results, the proposed HPANet-HFF obtained MAE and RMSE of 106.5 and 181.5 which is superior over other existing architectures.