Scalable and Resolution Data Analysis of Image and Video Compression using DL-CNNS Neural Network

Images and videos processing are becoming more and more crucial in digital interactions. For those interactions, deep learning has had great success in processing them. High-quality image and video footage is difficult to receive, distribute, reveal and compress due to the vast volume and advancements in resolution. The primary goal of this research was to develop a CNN based on DL for video and image compression to achieve good scalability and resolution. In this study, CNN was used to remove duplicate frames, while GAN and LSTM were utilized to identify minute changes and repeat single images and videos during compression. The DL-based CNN model's output vector calculates the degree of similarity between frames. The training process is stabilized and discrepancies between the generator output and the training aim are reduced using GAN. Repeated experiments on a range of videos and images with different sizes, durations, and quality levels demonstrated a significant resampling rate. In the end, the suggested approach achieves a 3% gain in PSNR, a 4.5% increase in SSIM, and a 50% decrease in RD cost during a video compression experiment—a considerable improvement over the performance of the current versatile video coding (VVC) technique in terms of PSNR, SSIM, compression ratio, and RD cost. The results of the image compression experiment include standard color images for testing using SSIM values (0.78–98), NPCR during compression (92–98%), and PSNR for image resolution (42–43%). When comparing the final product to the source video and image, there was often a 10% variation in quality and a size difference of more than half. Therefore, the proposed approach outperforms current approaches in terms of cost and compression ratio, according to simulation findings.