Method, device and medium for estimating diopter based on infrared eccentric photograph image

By constructing an array of LED beads and a multi-scale frequency domain decoupling network, the problems of lack of hardware physical structure perception capability and incomplete frequency domain feature decoupling in the existing technology are solved, achieving high-precision refractive power estimation, which is suitable for large-scale vision screening.

CN122271925APending Publication Date: 2026-06-26SUZHOU UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
SUZHOU UNIV
Filing Date
2026-03-26
Publication Date
2026-06-26

AI Technical Summary

Technical Problem

Existing technologies for estimating diopter using eccentric infrared images suffer from several drawbacks, including a lack of hardware physical structure perception capabilities, incomplete decoupling of frequency domain features, and weak multi-scale perception capabilities across large diopter spans. These issues lead to inaccurate measurements and poor robustness.

Method used

A diopter estimation method based on infrared off-center photography images is adopted. By constructing an array of LED beads, the hardware light source layout information is explicitly injected using residual convolutional layers and relative orientation feature modules. Combined with frequency domain wavelet transform and multi-scale gating aggregation modules, feature extraction and prediction are performed.

Benefits of technology

It improves the accuracy and robustness of refractive power detection, and can accurately predict spherical and cylindrical power in complex environments, making it suitable for large-scale vision screening.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122271925A_ABST
    Figure CN122271925A_ABST
Patent Text Reader

Abstract

This invention relates to the field of refractive error detection technology, and discloses a method, device, and medium for estimating refractive power based on infrared off-center photography images. The method uses infrared pupil images of each LED group in an array of LED beads when they are lit as feature images of each channel, forming a pupil feature image. This pupil feature image is input into the refractive error prediction model, where it is convolved through a residual convolutional layer to obtain a primary feature image. This primary feature image is then input into a relative orientation feature module to obtain a relative orientation feature image. The relative orientation feature image is then passed sequentially through multiple cascaded residual blocks to obtain residual features. These residual features are then passed through a global average pooling layer and fed into parallel and independent first and second fully connected layers to obtain predicted spherical and cylindrical refractive power values.
Need to check novelty before this filing date? Find Prior Art