Underwater image restoration via depth map and illumination estimation based on a single image

Area of Science:

• Computer Vision
• Image Processing
• Optical Engineering

Background:

• Traditional underwater image enhancement methods struggle due to simplified image formation models.
• The Akkaynak-Treibitz model offers improved robustness but is complex and underutilized.

Purpose of the Study:

• To develop an effective underwater image dehazing method using the revised Akkaynak-Treibitz model.
• To address limitations of previous methods in restoring underwater image quality.

Main Methods:

• Developed an underwater image depth estimation technique to generate scene depth maps.
• Utilized the revised model to estimate and remove backscatter based on pixel depth.
• Implemented an automatic color correction method to fix global color distribution.

Main Results:

• The proposed method effectively eliminates lightwave absorption and scattering using a single input image.
• Experimental results demonstrate superior contrast and color restoration compared to state-of-the-art techniques.
• The approach shows applicability across diverse real-world underwater environments.

Conclusions:

• The developed dehazing method significantly enhances underwater image quality.
• The integration of depth estimation and color correction with the revised model proves effective.
• This work promotes the adoption of more robust underwater image formation models for practical applications.