Optimized Video Compression and Artifact Reduction Using Convolutional Recurrent Kalman Filtering and Self-Adaptive Chaotic Enriched Grasshopper Optimization

Video compression methods are frequently employed to minimize the enormous amount of video data, but because they use lossy reduction, they also produce unappealing visual artefacts. In order to enhance the quality of the compressed images, this study offers a new framework for video compression called Convolutional Recurrent Kalman Filtering (CONV-RKF) model. That combines three essential elements: convolutional layers, a Recurrent Kalman Filter (RKF) and the Leaky ReLU activation function. Convolutional layers, which are at the heart of this system, are remarkably good at identifying complex spatial patterns within video frames and extracting relevant spatial information to create a base layer. The following combination of the Recurrent Neural Network (RNN) and Kalman Filter, an adaptive sequential method known as RKF for state estimation, allows the collection of temporal relationships in the video dataset. That makes it easier to forecast and estimate inter-frame motion accurately, which leads to a significant decrease in redundancy and an associated improvement in compression efficiency. In order to retain nonlinear fidelity throughout the compression process and to successfully mitigate problems like the “dying ReLU” phenomenon, the Leaky ReLU activation function plays a crucial role. To enhance the effectiveness of the CONV-RKF model, the hyperparameters are finetuned using the metaheuristic optimization model called Self-adaptive Chaotic Enriched Grasshopper Optimization (SCEGO) algorithm is introduced. The proposed CONV-RKF model is implemented using the MATLAB platform with the Peak Signal-to-Noise Ratio (PSNR) of 52.166 for the Quality factor (Q values) of 20, which is higher than the existing techniques.