Maintaining temporal coherence in video retargeting using mosaic-guided scaling.

Area of Science:

• Computer Vision
• Image Processing
• Multimedia Systems

Background:

• Video retargeting aims to adapt high-resolution videos to lower-resolution displays, often causing information loss.
• Maintaining spatio-temporal coherence is crucial for visual quality in retargeted videos.
• Existing methods struggle with temporal coherence due to camera and object motions.

Purpose of the Study:

• To propose a content-aware video retargeting method that ensures good temporal coherence.
• To address the challenge of maintaining temporal coherence in videos with camera and object motions.

Main Methods:

• Constructing a panoramic mosaic for each video shot.
• Aligning video frames to the panoramic mosaic.
• Deriving a global scaling map from the mosaic.
• Generating local scaling maps for individual frames, refined by spatial coherence constraints.

Main Results:

• The proposed method effectively maintains temporal coherence in retargeted videos.
• Achieves good visual quality, even in videos with significant camera and object motion.
• Demonstrates superior performance compared to existing video retargeting schemes.

Conclusions:

• Panoramic mosaic-guided video retargeting is a viable approach for preserving temporal coherence.
• The method offers a robust solution for high-quality video resizing under challenging motion conditions.