An underwater image sharpening method based on differential convolution and polarization fusion

By employing differential convolution and polarization fusion, the problem of insufficient color correction and structural detail restoration in underwater image sharpening was solved, achieving efficient sharpening and detail enhancement of underwater images, and improving target recognition and scene perception capabilities.

CN122415366APending Publication Date: 2026-07-17DALIAN MARITIME UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
DALIAN MARITIME UNIVERSITY
Filing Date
2026-06-05
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Existing underwater image sharpening methods struggle to simultaneously achieve color correction and structural detail restoration in complex and turbid water environments. They also fail to fully utilize polarization information and lack multi-scale detail restoration capabilities, resulting in limited improvement in underwater image sharpness and detail.

Method used

By employing differential convolution and polarization fusion, underwater RGB images and multi-angle polarization images are acquired, preprocessed, and spatially registered. Then, features are extracted using differential convolution branches and regular convolution branches, and polarization cues are generated by combining polarization parameters. Multi-scale feature interaction and fusion are performed, and finally, adaptive weighting and decoder reconstruction are carried out through a dual-branch enhancement network to achieve image sharpening.

Benefits of technology

It improves the ability to suppress scattering noise, express target contours and texture information in underwater images, enhances the ability to restore structures and reproduce colors in highly turbid environments, and improves the accuracy of underwater target recognition and scene perception.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122415366A_ABST
    Figure CN122415366A_ABST
Patent Text Reader

Abstract

The application discloses an underwater image sharpening method based on differential convolution and polarization fusion, comprising: spatially registering an RGB image and a polarization image of an underwater object; inputting the RGB image into a fusion differential convolution backbone network to cooperatively extract edge texture, color background and structure information of the image through differential convolution and conventional convolution, and obtaining embedded features; inputting the polarization image into a polarization coding branch to guide interactive fusion of the embedded features at different scales through multi-scale, and obtaining multi-modal fusion features; constructing an RGB spatial enhancement branch based on spatial distribution of the fusion features to enhance spatial response of a target region and significant texture; and obtaining a sharpened underwater image through a decoder and up-sampling reconstruction. The application effectively solves problems such as image detail blur, texture degradation and color shift in a high-turbidity underwater environment, enhances detail information and improves image sharpness while realizing color correction, and significantly improves underwater image quality.
Need to check novelty before this filing date? Find Prior Art