Underwater image background light optimization method and system based on autoencoder

By constructing an image quality evaluator and fitness function using an autoencoder, and optimizing the background light by combining Mahalanobis distance and reconstruction error, the problems of inaccurate background light estimation and unreasonable objective function design in underwater image restoration are solved, and accurate restoration of high-quality image features is achieved.

CN122134603APending Publication Date: 2026-06-02CHANGSHU INSTITUTE OF TECHNOLOGY
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
CHANGSHU INSTITUTE OF TECHNOLOGY
Filing Date
2026-05-06
Publication Date
2026-06-02

AI Technical Summary

Technical Problem

Existing underwater image restoration techniques suffer from problems such as background light estimation being susceptible to interference, unreasonable objective function design, and inaccurate image quality assessment, resulting in poor quality restored images.

Method used

An autoencoder-based approach is adopted. By constructing an image quality evaluator and fusing Mahalanobis distance and autoencoder reconstruction error, a fitness function is built. Heuristic optimization methods are used to search for the optimal background light, and multi-dimensional penalty terms are combined to optimize image features.

Benefits of technology

It significantly improves the accuracy and robustness of background light parameter estimation, ensuring that the features of the restored image are closely approximated by the high-quality image. It solves the optimization deviation problem caused by parameter errors in traditional methods and improves the restoration effect.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122134603A_ABST
    Figure CN122134603A_ABST
Patent Text Reader

Abstract

This invention discloses an underwater image background light optimization method and system based on an autoencoder. It constructs an image quality objective function based on an autoencoder, utilizing the autoencoder to learn the latent structural features of high-quality underwater images, allowing the objective function to optimize by ensuring the reconstructed image features closely match the high-quality image features. Image quality assessment is redefined as a geometric distance metric between the continuous distribution regions of the test image features and high-quality image features in the feature space. A feature deviation metric is constructed by weighted fusion of Mahalanobis distance and autoencoder reconstruction error, providing a precise and interpretable feature-level evaluation basis for the objective function. The reconstruction error output by the autoencoder is fused with Mahalanobis distance, and a fitness function is constructed by combining multi-dimensional constraint terms. This allows the optimization algorithm to search globally for the optimal background light value that best approximates the reconstructed image features to the high-quality image, solving the problem of local optima in background light estimation leading to optimization deviations from high-quality image features.
Need to check novelty before this filing date? Find Prior Art