Video Decolorization Using Visual Proximity Coherence Optimization.

Area of Science:

• Computer Vision
• Image Processing
• Digital Media

Background:

• Current video decolorization methods often adapt image techniques, leading to slow performance and inconsistent results.
• Preserving visual content while removing color information is crucial for various video applications.

Purpose of the Study:

• To develop an efficient and effective video decolorization framework that enhances temporal coherence and reduces processing time.
• To introduce a novel method that considers frame similarity and applies adaptive decolorization strategies.

Main Methods:

• A new metric, 'decolorization proximity,' was defined to quantify the similarity between adjacent video frames.
• Three distinct decolorization strategies were developed for frames based on their proximity levels (low, medium, high).
• A decolorization Gaussian mixture model was proposed to classify frames and assign appropriate strategies, optimizing for quality and coherence.

Main Results:

• The proposed framework significantly improves time efficiency compared to existing methods.
• Evaluations using qualitative, quantitative (color contrast preserving ratio, C2G-SSIM), and user studies demonstrate superior performance.
• A novel 'temporal coherence degree' metric confirmed enhanced consistency across decolorized video frames.

Conclusions:

• The developed video decolorization framework offers a significant advancement in efficiency and quality preservation.
• The method effectively balances the need for decolorization with maintaining temporal coherence in videos.
• This approach provides a robust solution for video decolorization challenges, outperforming current techniques.