Lens Model Generation Using Gaussian Blur Approximation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Camera lenses introduce aberrations such as geometric distortion, lateral chromatic aberration, and vignetting, leading to image blur that varies spatially across the image field, which existing technologies struggle to accurately model and correct.
Innovation Solution
A system and method for generating lens models based on reference images of a calibration pattern, using Gaussian approximations to create spatially variant blur patterns, allowing for the creation of global or local models that approximate blur across the entire image field or specific regions, effectively reducing image blur by applying these models to new images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If a lens model is generated to accurately represent spatially variant blur patterns, then image quality improvement is achieved, but computational resources are consumed
Solution Approach 1:
The patent uses Gaussian approximations with parameters (mean, variance) to model the blur pattern. By changing these parameters spatially across the image field, the system accurately represents the lens aberration without requiring complex computational models, thus improving image quality while controlling computational resource usage.
Solution Approach 2:
The system creates a simplified copy of the actual blur pattern using Gaussian functions. This approximation copy captures the essential characteristics of the spatially variant blur without the computational complexity of modeling every detail of the actual lens aberration, achieving quality improvement with reduced computational cost.
2Device complexity
If a global lens model is used to approximate blur across the entire image field, then model simplicity is maintained, but accuracy of local blur representation deteriorates
Solution Approach 1:
The patent applies different Gaussian approximation parameters at different locations across the image field. Each local region has its own blur characteristics modeled by adjusting the Gaussian parameters (mean and variance) to match the actual blur at that location, thereby achieving accurate local representation while maintaining overall model simplicity.
Solution Approach 2:
The image field is segmented into multiple regions, each with distinct blur characteristics. The global model is divided into multiple local Gaussian approximations, one for each region. This segmentation allows the system to maintain simplicity at the global level while achieving accuracy at the local level through region-specific parameter adjustment.
3Measurement precision
If multiple local lens models are used to approximate blur in each local region, then local blur representation accuracy is improved, but device complexity increases
Solution Approach 1:
Instead of using completely different models for each local region, the patent changes only the parameters (mean and variance of Gaussian functions) while maintaining the same basic Gaussian model structure. This approach improves local accuracy through parameter optimization without significantly increasing the fundamental model complexity.
Solution Approach 2:
Each local region is modeled with a Gaussian function whose parameters are optimized for that specific region's blur characteristics. This local quality approach ensures accurate representation in each region while using the same simple Gaussian framework across all regions, preventing exponential complexity growth.
Data Source
AI summary
Techniques are disclosed relating to lens modeling. In one embodiment, a lens model may be generated based on reference images of a pre-determined, known geometric pattern. The lens model may represent a spatially variant blur pattern across the image field of the lens used to capture the reference images. In one embodiment, the lens model may include Gaussian approximations of the blur that may minimize the difference between a location within a reference image and a corresponding location of a pre-determined, known geometric pattern. In one embodiment, the generated lens model may be applied to deblur a new image.