Row Noise Filtering in CMOS Image Sensors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Row correlated noise in image data from image sensors, particularly in CMOS image sensors, remains a challenge due to power supply noise injection and limitations in matching reference samples with exact loading, bandwidth, and sampling instants, leading to visible noise even after power supply noise reduction and reference generation correction.
Innovation Solution
A method involving the application of a spatial filter to row averages to reject high-frequency signals, calculating a correction value from the difference between actual and desired row averages, and applying this correction to the data, which reduces row noise while minimizing image blurring.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If a 2D spatial filter is used to remove row noise, then row noise is reduced, but image blurring increases
Solution Approach 1:
The patent segments the image processing into two distinct stages: first computing row averages to capture noise patterns, then applying the spatial filter only to these averages. This separation allows noise reduction without directly filtering the original image data, thereby avoiding blurring while maintaining image quality.
Solution Approach 2:
The row average serves as an intermediary representation that captures the noise characteristics without containing the fine detail information of the original image. By filtering this intermediate representation rather than the original image data, the patent achieves noise reduction while preserving image sharpness.
2Object-affected harmful factors
If traditional 2D spatial filtering is applied to remove row noise, then noise is reduced, but processing complexity increases
Solution Approach 1:
The patent divides the complex 2D spatial filtering operation into two simpler operations: computing row averages (1D operation) and then filtering only these averages. This segmentation reduces the computational complexity compared to applying a full 2D filter to the entire image.
Solution Approach 2:
The patent extracts only the noise-containing information into the row averages, separating it from the image content. By filtering only this extracted noise component rather than the entire image data, the patent reduces processing complexity while maintaining effectiveness.
Data Source
AI summary
Apparatus and a method for processing image data where row data is received corresponding to a target row of pixels and to one or more reference rows of pixels. A target row average is generated based on the target row data and a reference row average is generated for each of the one or more reference rows based on each reference row's respective row data. A row correction value is generated based on the target row average of the target row and the reference row average of the one or more reference rows. Corrected target row data is generated by applying the row correction value to the target row data.