A Variational Approach to Video Registration with Subspace Constraints

Area of Science:

• Computer Vision
• Image Processing
• Deformable Object Analysis

Background:

• Non-rigid video registration is challenging due to deformable objects.
• Accurate optical flow computation is crucial for understanding object motion.

Purpose of the Study:

• To develop a robust and efficient method for non-rigid video registration.
• To improve the accuracy and temporal consistency of optical flow estimation for deformable objects.

Main Methods:

• Exploiting high correlation between 2D trajectories of points on non-rigid surfaces.
• Using a low-rank motion basis as a subspace constraint within a variational framework.
• Developing a novel optimization scheme for vector-valued images using dualization.

Main Results:

• Achieved temporally consistent optical flow for deformable objects.
• Significant improvements in registration results for color images.
• Outperformed state-of-the-art optical flow and non-rigid registration algorithms.

Conclusions:

• The proposed low-rank subspace constraint provides effective trajectory regularization for optical flow.
• The novel optimization scheme extends the approach to color images, enhancing performance.
• The new benchmark dataset facilitates rigorous evaluation of non-rigid optical flow algorithms.