A Comprehensive Framework for Video Resolution Enhancement: using Computer Vision and Deep Learning Techniques

The paper introduces a computer vision based four-phase framework for enhancing video resolution by removing motion blur and motion based temporal aliasing. Beginning with stabilization using the Lucas-Kanade method, a stable foundation is created. Subsequent stages involve deblurring through Gaussian blurring and Wiener deblurring, addressing temporal aliasing with bicubic interpolation. To further boost resolution a final denoising stage using Non-Local Means method reduces noise while preserving details in a video. To overcome the limitations of computer vision based model we have designed a deep learning approach with the U-Net architecture employed for large motion frame interpolation. U-Net enhances spatial features and even uses Super Resolution Convolutional Neural Network (SRCNN) for deblurring of video and making the video more sharper. This innovative methodology aims to redefine video resolution, with potential applications in surveillance, video analytics and content creation.