Published Paper Details:

This paper addresses challenges in low-light image enhancement, crucial for applications like autonomous driving and surveillance. Existing methods, including traditional ones like histogram equalization and newer deep learning approaches, have limitations such as unnatural contrast and noise. The proposed Multi-Level Network Fusion (MFIE-Net) algorithm integrates pre-processing, image fusion, and convolutional neural network-based feature extraction. It employs multi-level network fusion to mitigate down sampling-induced feature loss and incorporates a comprehensive loss function for overall quality enhancement. The Retinex method decomposes images into illumination and reflectance components, addressing uneven lighting conditions. Multi-Scale Network Fusion models local context in multiscale network integration, using weighted features at different scales. The paper explores neural methods' applications, including Generative Adversarial Networks (GANs), Neural Style Transfer, Deep Reinforcement Learning, and neural networks in animation, interactive graphics, image enhancement, and 3D graphics. The conclusion emphasizes the importance of image enhancement, introducing three prominent methods: Retinex, Multi-Scale Network Fusion, and Local Context Modeling. Each method has unique advantages, and the choice depends on specific image characteristics and enhancement goals. The research contributes by introducing the innovative MFIE-Net and providing insights into neural methods across diverse computer graphics domains.