Publication

Abstract :

A novel general framework to create an accurate algorithm for video enhancement tasks, including deblurring, and denoising, is presented in this paper. The understanding that high-quality video enhancement requires precision rather than the density of pixel flows is crucial for the framework. The majority of earlier research uses the opposite strategy with the estimation of dense (per-pixel) flows using computationally expensive algorithms. A unique mesh-based flow estimating algorithm is suggested here instead to estimate the flow information. An automated subparametric curved triangular mesh generator for video frames in Python from Gmsh as part of the development of the suggested framework is created. At each nodal point, the optical flow values are then calculated. Using the autoencoder technique, denoising is applied to each video frame. Each frame’s video quality is improved by using optical flows and CNN algorithms. A general architecture that integrates the flows in a plug-and-play manner with various task-specific layers, building on top of the flow estimation, is presented. The quality of algorithms created using the proposed framework is improved by 0.42 dB to 6.70 dB compared to computing techniques.