High capacity Reversible Data Hiding scheme using prediction tuning model

With the growing importance of data security and information hiding, techniques of data hiding in images have become widespread. Existing reversible data hiding techniques face the problem of errors due to inaccurate results of prediction models as well as the scope of improvement in terms of embedding capacity, which decreases the efficiency of security of sensitive data. In this paper, we propose a hyperparameter beta ((3) for combining two regression models trained with different data. The first regression model is trained over the entire image and the second regression model with a limited block of surrounding pixels in neighbouring regions of pixels in the image. For optimal results, parameters are tuned using Particle Swarm Optimization (PSO), and we have validated our results on standard test images. This novel technique outperforms the past techniques by improving data hiding in terms of parameters like embedding capacity (embedding rate), eliminating the errors from wrong predictions while keeping the visual quality of images preserved.