Effective reference frame selection for video colorization: an analysis of manual and automatic selection approaches

Abstract

This paper investigates the impact of reference frame selection on the quality of colorization in videos. A comprehensive analysis of three different video datasets was conducted to evaluate several manual and automatic reference frame selection approaches. Our findings reveal that automatic selection approaches outperform the conventional first or last frame selection approach. Additionally, we propose a novel effective multi-reference frame selection framework that enhances colorization accuracy by selecting an efficient set of reference frames based on the spatio-temporal correspondences of video frames. We believe this is the first study to provide an extensive analysis of the performance of various reference frame selection approaches in video colorization, offering significant implications for future work in this field.