High-Resolution Eyelash Removal in Ultra-Wide-Field Fundus Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for eyelash removal in ultra-wide-field fundus images fail to effectively retain high-resolution details and under-eye image information, limiting the accuracy of retinal disease diagnosis and screening.
Innovation Solution
A two-stage architecture comprising a dual super-resolution learning network for eyelash segmentation and a large mask inpainting model using Fast Fourier Convolutions to virtually remove eyelashes, preserving image resolution and detail.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If rotating and cropping techniques are applied to avoid eyelash regions, then eyelash artifacts are reduced, but detail outside the eyelash region and under-eye image information are lost
Solution Approach 1:
The patent extracts and removes only the eyelash region from the fundus image while preserving the rest of the image content. The method segments the eyelash region and applies inpainting to reconstruct only the affected areas, rather than cropping entire regions, thus maintaining under-eye image information and other critical retinal details.
Solution Approach 2:
The patent divides the image processing into distinct segments: eyelash region segmentation, mask generation, and selective inpainting. This segmentation allows precise removal of eyelashes while preserving surrounding structures, enabling detailed retention of under-eye regions and other important fundus areas.
2Device complexity
If traditional image processing methods such as edge detection and thresholding are used for eyelash segmentation, then processing simplicity is maintained, but segmentation accuracy and reliability are insufficient for medical diagnostic applications
Solution Approach 1:
The patent replaces traditional mechanical image processing methods (edge detection, thresholding) with a deep learning-based segmentation model. This substitution enables significantly improved segmentation accuracy and reliability for medical diagnostic applications, while the overall system remains integrated and coherent.
Solution Approach 2:
The patent transitions from simple binary thresholding parameters to complex multi-parameter deep learning models that consider spatial, textual, and contextual information. This parameter transformation enables accurate segmentation of eyelash regions even in complex fundus image conditions, achieving reliability suitable for medical diagnostics.
3Manufacturing precision
If eyelash removal methods are applied to retain high-resolution details, then diagnostic quality is improved, but processing time and computational resources increase
Solution Approach 1:
The patent performs preliminary actions by pre-training the deep learning segmentation model and preparing the inpainting system in advance. The model is pre-trained on annotated fundus images, enabling rapid inference during actual diagnostic processing. This preliminary preparation significantly reduces processing time while maintaining high-resolution output quality.
Solution Approach 2:
The patent applies partial inpainting only to the eyelash regions identified by the segmentation model, rather than processing the entire image. This selective approach reduces computational burden and processing time while maintaining high resolution and detail in the critical areas where eyelashes were present.
Data Source
AI summary
A method allows virtual eyelash removal in high-resolution ultra-wide-field fundus images. A deep learning approach removes eyelash artifacts from UWF fundus images, enhancing clinical utility by providing clearer and accurate images for analysis and diagnosis. A high-resolution segmentation and image inpainting model obtains a super ultra-wide-field fundus image that virtually removes eyelashes and retains high resolution and high realism from a UWF fundus image with occluding eyelashes.