Rectangling and enhancing underwater stitched image via content-aware warping and perception balancing

Area of Science:

• Computer Vision
• Image Processing
• Marine Technology

Background:

• Underwater images suffer from limited field-of-view and poor visual perception due to scattering and absorption.
• Existing image stitching methods produce irregular boundaries, and deep learning approaches lack reliable references, causing distortions.

Purpose of the Study:

• To propose a novel framework, Underwater Wide-field Image Rectangling and Enhancement (UWIRE), for rectifying and enhancing single underwater stitched images.
• To address limitations in field-of-view, visual perception, and boundary distortions in underwater imagery.

Main Methods:

• The UWIRE framework includes an R-procedure for rectangling irregular boundaries using shape resizing and mesh-based warping with complementary optimization.
• An E-procedure enhances images via parameter-adaptive correction and an attentive weight-guided fusion for color, contrast, and texture.

Main Results:

• The R-procedure ensures natural appearance with minimal distortion by optimizing boundary, structure, and content.
• The E-procedure balances information distribution across channels for improved image quality.
• Comprehensive experiments show UWIRE outperforms state-of-the-art methods in quantitative and qualitative evaluations.

Conclusions:

• The proposed UWIRE framework effectively rectifies and enhances underwater stitched images using a single input.
• UWIRE offers a robust solution for improving underwater image quality without requiring external references or complex setups.