Neural Defocus Blur Reshaping from Multi-Pupil Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods fail to accurately reproduce defocus blur for pupils larger than the optical system, leading to increased noise and decreased spatial resolution due to vignetting and uneven weight adjustments, especially with annular patterns from aspheric lenses.
Innovation Solution
An image processing method using deep learning to reshape defocus blur by inputting images captured through different pupils of an optical system, employing a multilayered neural network to combine and reshape defocus blur into desired shapes like flat circles and Gaussian distributions, while suppressing unwanted effects like vignetting and multiple blurs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If the pupil in the optical system is subdivided to reduce the influence of annular patterns, then the quality of defocus blur is improved, but the spatial resolution decreases and noise increases in each disparity image
Solution Approach 1:
The pupil is divided into multiple sub-pupils to capture multiple disparity images, each with reduced annular pattern influence. The segmentation allows the system to process and combine images from different pupil regions, achieving better overall defocus blur quality while managing the trade-off with spatial resolution through intelligent combination strategies.
Solution Approach 2:
Multiple disparity images captured from different sub-pupils are combined using weight adjustment to produce a final image with improved defocus blur quality. The merging process integrates information from multiple sources to compensate for the spatial resolution loss and noise increase in individual images.
2Manufacturing precision
If the pupil in the optical system is subdivided to reduce the influence of annular patterns, then the quality of defocus blur is improved, but the noise increases in each disparity image
Solution Approach 1:
The pupil is divided into multiple sub-pupils to capture multiple disparity images, each with reduced annular pattern influence. The segmentation allows the system to process and combine images from different pupil regions, achieving better overall defocus blur quality while managing the trade-off with spatial resolution through intelligent combination strategies.
Solution Approach 2:
Multiple disparity images captured from different sub-pupils are combined using weight adjustment to produce a final image with improved defocus blur quality. The merging process integrates information from multiple sources to compensate for the spatial resolution loss and noise increase in individual images.
3Shape
If existing methods adjust weights when combining disparity images, then the defocus blur shape is controlled, but the defocus blur corresponding to a pupil larger than the optical system cannot be reproduced
Solution Approach 1:
The system changes the parameter of pupil size by capturing images from multiple sub-pupils and combining them with adjusted weights. This parameter change approach enables the reproduction of defocus blur corresponding to a larger effective pupil than the physical optical system, achieving both shape control and enhanced adaptability.
Data Source
Figure 1
Figure 2~3
Figure 4A~4B
AI summary
An image processing method includes the steps of acquiring (S101, S301) a first image obtained by imaging an object space through a first pupil in an optical system, and a second image obtained by imaging the object space through a second pupil different from the first pupil in the optical system, and generating (S102, S307) a blur reshaped image in which a blur caused by a defocus is reshaped based on the first image and the second image using a neural network.