Deep image enhancement for ill light imaging | JoVE Visualize

Area of Science:

Computer Vision
Image Processing
Deep Learning

Background:

Capturing natural scenes under adverse lighting (low light, back-lit, over-exposed) leads to simultaneous over- and under-exposure, complicating image processing.
Existing methods using multiple sensors or handcrafted parameters have limited capacity, often failing to address specific scene conditions and maintain visual smoothness, color, and contrast.
Single image sensors struggle to capture satisfactory quality in ill-lit environments, posing challenges for subsequent analysis and application.

Purpose of the Study:

To propose a novel deep learning-based image enhancement method for color images captured under ill lighting conditions.
To address challenges of illumination blindness, structure degradation, and noise while preserving visual smoothness, color, and contrast.
To develop a robust model capable of handling diverse and extreme lighting scenarios.

Main Methods:

Input images are decomposed into reflection and illumination maps using a proposed layer distribution loss network.
A knowledge-based adaptive enhancement constraint tunes hidden degradation in reflection and illumination maps.
A Repair-Net with a total variation operator optimizes local consistency and guides image gradients for enhanced reconstruction.

Main Results:

The proposed method effectively solves illumination blindness and structure degradation problems by decomposing images into reflection and illumination components.
The model successfully balances smoothness, reduces noise, and mitigates over- and under-enhancement issues.
Experimental results on a new dataset under very low exposure and ill illumination conditions demonstrate superior performance in preserving structural and textural details compared to state-of-the-art methods.

Conclusions:

The developed deep image enhancement method offers superior performance for images captured under challenging lighting conditions.
The approach demonstrates practical utility for future visual applications requiring high-quality image data from adverse environments.
The method's ability to preserve details and maintain visual integrity makes it a significant advancement in low-light image processing.