Enhancing the Compressed Video Quality through a Neural Network Approach
Karthick Panneerselvam, K. Rajalakshmi, P. Sornalatha · 2025
Video compression mainly focuses on reducing the file size of videos with little loss of quality, but most video compression algorithms on the market do not focus on clarity. Data compression is increasingly being considered as a necessary issue in larger engineering communities. Data compression using neural networks has been the main topic of some related research. The main objective of this study would be to apply the video in an amazing way with the help of an Enhanced Generative Adversarial Network (EGAN). EGAN was better to manage and produced better consistent results with fewer artifacts than other Neural Networks (NN). Deep neural networks (DNNs), recurrent neural networks (RNNs), generational adversarial networks (GANs), and various autoencoder (AE) variants have all been used for various compression methods. It would be a potential option for incredible movies. Based on significantly reduced sample input films, the proposed neural network was able to maintain high-frequency information and generate pleasurable sensory videos. A Residual Dense Block (RDB) is the sort of block utilized in the design. In comparison to previous state-of-the-art approaches, the comprehensive Mean-Opinion-Score (MOS) test, and the Video Multi-Method Assessment Fusion (VMMAF) test, reveal extremely substantial increases in terms of human experiential accuracy. The researchers are also proposing a new data pre-processing method based on the Deep Recurrent Neural Network (DRNN) that has been combined with two leading-edge coding standards, and the Adaptable Video Coding (AVC). The results show consistent encoding benefits across all sequences examined at multiple spatial frequencies, with an average bit rate savings of 4.0 percent relative to AVC.