Robust disparity estimation based on color monogenic curvature phase

Area of Science:

• Computer Vision
• Image Processing
• 3D Reconstruction

Background:

• Disparity estimation is vital for 3D tasks like virtual reality and robot navigation.
• Conventional methods fail with large illumination variations and noise due to reliance on brightness constancy.
• Accurate disparity maps are essential for reliable 3D environment reconstruction.

Purpose of the Study:

• To develop a robust disparity estimation method for binocular images.
• To overcome limitations of conventional approaches in challenging lighting and noisy conditions.
• To improve the accuracy of disparity maps in real-world scenarios.

Main Methods:

• Proposed a color monogenic curvature phase model using quaternion representation for local feature description.
• Developed a multiscale framework integrating color monogenic curvature phase with mutual information.
• Utilized both indoor and outdoor datasets with significant brightness variations for testing.

Main Results:

• The proposed approach demonstrated robust performance in estimating disparities.
• Effective handling of large illumination changes and serious noisy contamination was achieved.
• Experimental results confirmed the method's superiority over conventional techniques in adverse conditions.

Conclusions:

• The color monogenic curvature phase and mutual information coupling offers a powerful solution for robust disparity estimation.
• This method significantly enhances the accuracy of disparity maps under challenging visual conditions.
• The approach is suitable for various applications requiring precise 3D information from binocular images.