Robust Matching Cost Function for Stereo Correspondence Using Matching by Tone Mapping and Adaptive Orthogonal...

Area of Science:

• Computer Vision
• Image Processing
• Stereo Vision

Background:

• Real-world stereo images suffer from radiometric differences like exposure and lighting variations.
• Severe radiometric distortion can lead to nonlinear intensity changes, challenging stereo matching.

Purpose of the Study:

• To propose a robust and accurate stereo matching cost function resilient to radiometric variations.
• To introduce an adaptive orthogonal integral image for improved stereo image analysis.

Main Methods:

• Developed an adaptive orthogonal integral image, an enhancement of the original orthogonal integral image.
• Proposed a matching cost function utilizing tone mapping and the adaptive orthogonal integral image.
• Enabled adaptive support region construction for pixel-wise robustness within object boundaries.
• Developed automatic parameter estimation techniques for the proposed function.

Main Results:

• The proposed matching cost function achieved error reductions from 21.51% to 15.73% on the Middlebury dataset.
• Error rates decreased from 15.9% to 10.85% on the Kitti dataset compared to using the orthogonal integral image.
• Demonstrated superior performance over census transform, local binary patterns, and normalized cross-correlation methods.

Conclusions:

• The novel matching cost function effectively handles locally nonlinear intensity distortion in stereo images.
• The adaptive orthogonal integral image facilitates robust matching across object boundaries.
• Experimental results confirm the proposed method's superiority for stereo matching under radiometric variations.