Image Sensor Fixed Pattern Noise Reduction via Center-Weighted Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing technologies fail to effectively reduce fixed patterns generated by repeating cycles in image sensors without compromising image detail or quality.
Innovation Solution
An image processing device that uses a filter with coefficients weighted towards the center and a size one larger than the pixel configuration, ensuring equal sums of coefficients for pixels in identical positional relationships, thereby reducing fixed patterns while preserving image detail.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If a filter with center-weighted coefficients is applied while shifting the pixel group, then noise reduction is improved, but fixed patterns become more notable and image quality deteriorates
Solution Approach 1:
The patent applies different processing approaches to different regions based on their characteristics. Specifically, it identifies fixed pattern regions and applies targeted correction filters only where needed, while preserving detail in other regions. This local differentiation resolves the contradiction by reducing noise in fixed pattern areas without blurring important image details elsewhere.
Solution Approach 2:
The patent dynamically adjusts filter parameters based on the detected fixed pattern characteristics. By changing the filter coefficients and processing parameters according to the specific fixed pattern situation, it achieves effective noise reduction while maintaining image quality. The parameter adaptation allows the system to optimize between noise reduction and detail preservation.
2Reliability
If a filter with uniform coefficients is applied, then fixed patterns are reduced, but image detail is lost and quality deteriorates
Solution Approach 1:
The patent applies uniform coefficient filters specifically to fixed pattern regions while using different processing for other areas. This localized application ensures that fixed patterns are reduced without unnecessarily blurring important image details in regions where fixed patterns are not present.
Solution Approach 2:
The patent segments the image processing into different regions: fixed pattern regions and non-fixed pattern regions. Different filter strategies are applied to each segment, with uniform coefficients used only where needed for fixed pattern reduction, thereby preserving image detail in other segments.
3Reliability
If a larger filter size (6x6 pixels) is used to match the basic array pattern, then fixed patterns are reduced, but image detail and quality deteriorate
Solution Approach 1:
The patent applies the 6x6 filter specifically to regions containing fixed patterns rather than uniformly across the entire image. This localized application reduces fixed patterns while preserving image detail in areas where the large filter would otherwise cause excessive blurring.
Solution Approach 2:
The patent segments the image into fixed pattern regions and non-fixed pattern regions, applying the 6x6 filter only to the former. This segmentation strategy allows effective fixed pattern reduction without sacrificing image detail in regions where the larger filter size would be detrimental.
Data Source
Figure 1A~1B
Figure 2A~2B
Figure 3
AI summary
A method for image processing according to one aspect of the present invention includes: a step of acquiring an image taken by imaging means including an image sensor having pixel configuration with repeating cycles of M × N (M, N: integers of 2 or more) pixels; (a) a step of setting a target pixel in the acquired image and extracting K × L (K, L: integers of M<K and N<L) pixels based on the target pixel; (b) a step of calculating a pixel value of the target pixel with an operation with use of a filter which has a K × L filter size and which has specified filter coefficients arrayed therein; and a step of repeatedly executing the step (a) and the step (b) while moving the target pixel one pixel at a time with respect to the acquired image.