Image Correction via Deformation-Aware Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Image quality deterioration due to geometric transformation, such as that caused by fisheye lens distortion, is not accurately corrected, often resulting in insufficient or excessive correction.
Innovation Solution
An image processing method that acquires information about the deformation amount of the image and uses a machine learning model, specifically a neural network, to generate an estimated image that corrects image quality deterioration by applying the deformation information to the image processing system.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a fixed deformation amount is used for correction irrespective of geometric transformation applied, then the correction process is simple, but insufficient correction or excessive correction occurs depending on the geometric transformation
Solution Approach 1:
The patent applies dynamics by making the deformation amount variable rather than fixed. The system dynamically adjusts the deformation amount based on the actual geometric transformation applied to the image, allowing the correction parameter to adapt to different transformation conditions and achieve accurate correction without being overly simplistic.
Solution Approach 2:
The patent changes the parameter of deformation amount from a fixed value to a variable that depends on the geometric transformation. By calculating the actual deformation amount based on the transformation applied, the system optimizes the correction parameter to match the specific image processing conditions, resolving the contradiction between simplicity and accuracy.
2Area of stationary object
If geometric transformation is applied to correct fisheye lens distortion, then wide-range imaging is achieved, but image quality deteriorates in regions with large deformation amounts
Solution Approach 1:
The patent applies local quality by differentiating the correction approach based on the deformation amount in different image regions. Regions with large deformation amounts receive different correction treatment compared to regions with small deformation amounts, allowing the system to maintain image quality in critical areas while preserving the wide-range imaging capability.
Solution Approach 2:
The patent changes the correction parameter (deformation amount) based on the local characteristics of each image region. By calculating and applying region-specific deformation amounts, the system optimizes image quality in areas with large transformations while maintaining the overall wide-field imaging benefit.
3Measurement precision
If machine learning model is used to correct image quality deterioration, then correction accuracy improves, but the system requires additional information about deformation amount
Solution Approach 1:
The patent applies preliminary action by calculating and storing the deformation amount information during the image processing pipeline before the machine learning correction step. This pre-computation of deformation data prepares the necessary input for the machine learning model, enabling accurate correction without adding complexity to the overall system architecture.
Solution Approach 2:
The patent introduces deformation amount information as an intermediary parameter that bridges the geometric transformation process and the machine learning correction process. This intermediary data structure allows the machine learning model to receive precise input about the transformation applied, enabling accurate correction without requiring the model to directly analyze transformation parameters.
Data Source
AI summary
An image processing method includes acquiring a second image obtained by applying geometric transformation to a first image, acquiring information about a deformation amount of the first image in the geometric transformation, and generating a third image based on the second image and the information about the deformation amount.