Multi-View Disparity Estimation Using the Gradient Consistency Model.

Area of Science:

• Computer Vision
• Image Processing
• Computational Photography

Background:

• Variational methods for disparity estimation often use linearized brightness constancy, limiting accuracy in non-smooth areas.
• Current methods rely on predefined schedules for data inclusion, which can be suboptimal.

Purpose of the Study:

• To propose a novel Gradient Consistency Model for disparity estimation.
• To improve the accuracy and convergence rate of variational disparity estimation methods.
• To introduce a self-scheduling mechanism that adapts to image data validity.

Main Methods:

• Utilized Gradient Consistency information to assess linearization validity in disparity estimation.
• Developed an analytically inspired Gradient Consistency Model that penalizes spatial gradient mismatches between views.
• Implemented a self-scheduling approach where data term weights evolve during the algorithm's progression.

Main Results:

• The Gradient Consistency Model demonstrated superior performance compared to standard coarse-to-fine schemes.
• Outperformed the progressive inclusion of views approach in both convergence rate and accuracy.
• Effectively determined data term weights based on gradient consistency, avoiding reliance on external schedules.

Conclusions:

• The Gradient Consistency Model offers a more robust and accurate approach to variational disparity estimation.
• Self-scheduling based on gradient consistency eliminates the need for tuned or learned schedules.
• This method advances disparity estimation techniques in computer vision.