Year: 2024
Type: conference-paper
Abstract: Deep learning has achieved remarkable success in video colorization problems. Reference-based video coloring problem is an important topic. In this paper, we propose a reference-based video colorization model that distinguishes itself from conventional colorization approaches. Unlike exemplar-based colorization methods that focus on leveraging color information from the reference frame, our model addresses the inherent limitation of the reference frame potentially lacking essential color details. To tackle this issue, we introduce the AB chrominance point to supplement the missing color information in the reference frame, yielding more precise colorization results. While traditional models often employ one-way or simple bidirectional optical flow, we use a bidirectional optical flow propagation network to make full use of the information from the warping images and increase the temporal consistency. Experimental results substantiate the efficacy of our approach, showcasing superior performance compared to state-of-the-art methods on benchmark datasets.
Authors: Yizhang Yang, Jianxiu Jin, Ying Huang, Kailing Guo, Xiangmin Xu
Institution: South China University of Technology
Language: English
Topic: Image Enhancement Techniques
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4399418399
