Asymmetric Stereo High Dynamic Range Imaging with Smartphone Cameras.

Area of Science:

• Computer Vision
• Image Processing
• Computational Photography

Background:

• Stereo high dynamic range imaging (SHDRI) provides temporally stable HDR images, outperforming traditional bracketing methods.
• Existing SHDRI solutions struggle to leverage asymmetric lenses common in smartphones.
• Single-image HDR methods often sacrifice accuracy.

Purpose of the Study:

• To develop a novel method for single-shot asymmetric SHDRI using deep learning.
• To address the limitations of current SHDRI techniques in utilizing asymmetric smartphone lenses.
• To improve the robustness and accuracy of HDR fusion.

Main Methods:

• A reference-based deep learning approach was employed for single-shot asymmetric HDR fusion.
• The method was designed to handle differences in aperture and image signal processing pipelines.
• Utilized the unique characteristics of asymmetric lenses found on modern smartphones.

Main Results:

• The proposed system successfully achieved single-shot asymmetric HDR fusion.
• Demonstrated superior robustness to variations in aperture and image signal processing pipelines compared to existing solutions.
• Maintained accuracy in HDR reconstruction despite lens asymmetry.

Conclusions:

• The developed deep learning method offers a viable and robust solution for SHDRI using asymmetric smartphone lenses.
• This advancement enables higher quality HDR imaging on mobile devices.
• The approach represents a significant step forward in computational photography for consumer electronics.