Video Super-Resolution Using Machine Learning

Please enable javascript in your browser.

Ashoori, Mohammad Hossein | 2022

388 Viewed

Type of Document: M.Sc. Thesis
Language: Farsi
Document No: 55072 (05)
University: Sharif University of Technology
Department: Electrical Engineering
Advisor(s): Amini, Arash; Marvasti, Farrokh
Abstract:
Super-resolution means increasing the resolution so that the quality improves. This is defined for both image and video. In this regard, machine-learning based methods, especially convolutional neural networks, have shown great potential in recent years. Finding the right structure that can deliver high speed and accuracy is the key to solving the super-resolution problem.Despite the myriad of methods for image super-resolution, less attention has been paid to its generalization to video. This generalization should be such that more detail is created in the output using adjacent frames.In this dissertation, the existing methods for image and video super-resolution are reviewed, and then a new structure is proposed. By adjusting the hyperparameters of the proposed structure, three networks with a number of different parameters are trained and their quantitative and qualitative results are compared with existing methods.Although according to some quantitative criteria, the proposed networks do not provide better results than the existing methods, in terms of output image quality, they have acceptable performance
Keywords:
Super Resolution ; Convolutional Neural Network ; Machine Learning ; Separable Convolutional Layer ; Batch Normalization (BN)Layer Network ; Image Quality