OLD VIDEO REJUVENATION USING MACHINE LEARNING TECHNIQUES

• Main Author: Adrian, Dicky
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/48308
• Institution: Institut Teknologi Bandung

Old video rejuvenation can be done by improving video’s FPS and coloring each frame of the video. There are already some conventional methods used to do these task such as using motion blur and coloring each frame one by one. Although these methods produce a good result, by using motion blur, the video will lose a lot of details, and coloring each frame manually will take a long time to complete. From these concerns, in this final project the goal is to use machine learning techniques to accomplish these tasks. To achieve this project’s goals, we need to implement the best techniques on each components needed for video rejuvenation. For that purpose, first some experiments are done quantitatively, qualitatively and also for load performance of each model. Quantitative experiments are done by comparing quantitative metrics from generated frame with the original frame. As for qualitative experiments, the frames generated from the techniques tested will be compared directly by human eyes. Lastly load performance testing are done by measuring how fast each technique generate a new frame based on the inputs. Based on our experiments and implementation, the application can do old video rejuvenation quite well. In our testing, we got an average of PSNR of 30.07 and average of SSIM of 0.85 from approximately 100,000 frames. With these measurements of PSNR and SSIM, human eye cannot tell any differences between frames generated by machine learning and the original frames.