Master of Science, M.Sc., Student in Big Data Analytics
San Diego State University
rpalafler1@gmail.com
Successfully Defended and Unanimously Approved on March 17th, 2023 to the Thesis Committee and the General Public.
The Defense fully explains the preliminary information needed for handling image and video data, the methodologies of frame interpolation, resolution upsampling, parallel processing, bootstrapping, and statistical inference, and includes an interpretation of the results with a live demo of the Python program.
The purpose of this Thesis is to provide an automated, user-driven, and scalable programmatic approach to video remastering. Supported by a fully customizable and working program built in Python, this Thesis demonstrates methods for handling large image (array-based) data through parallel processing—commonly referred to as “multiprocessing”—by (1) doubling the original video’s framerate and (2) upsampling the video’s existing resolution to a higher resolution with additional enhancements made to image quality.
Suitable for remastering any kinds of videos including films, movies, television shows, video game cutscenes, personal videos, and historical footage, the methods explained in this Thesis provide baseline results to compare with more advanced neural network models requiring prior training and frequent updating.
Combined with a suite of statistical metrics measuring image quality between the original and "remastered" frames, the baseline results derived from the methods shown in this Thesis can be quantitatively and visually used to determine whether a more advanced video remastering model sufficiently improves upon them or not.