CMOS Row Noise Reduction via Shielded Pixel Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing row noise reduction methods in CMOS image sensors are ineffective in eliminating vertical non-uniformities and are prone to offset estimation errors due to bad pixels, requiring significant memory and processing time, and are not suitable for sensors with Color Filter Arrays (CFAs) without interpolation.
Innovation Solution
Determining row noise offset using statistics from dark or shielded pixels, which reduces outlier effects and eliminates the need for edge detection routines, allowing for more robust noise reduction before color interpolation and reducing memory requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If localized row-by-row correction is used to reduce row noise, then row noise is corrected, but vertical non-uniformity in sensor response is not eliminated and offset estimation errors increase due to bad pixels
Solution Approach 1:
The patent segments the pixel array into multiple blocks, each containing both active and dark pixels. This segmentation allows independent offset estimation for each block, isolating the influence of bad pixels to local regions rather than affecting the entire row. The segmented approach enables more reliable statistics to be collected from dark pixels within each block, improving offset estimation accuracy while maintaining row noise correction effectiveness.
Solution Approach 2:
The patent introduces dark pixels as intermediary elements that serve as a reference for offset estimation. These dark pixels, which are shielded from light, provide a stable baseline signal that is not affected by scene content or bad pixels in the active area. By using dark pixel statistics as an intermediary reference, the method achieves more accurate offset estimation without being influenced by outliers in the active pixel data.
2Reliability
If several rows of pixels are stored in memory for statistical analysis, then row noise correction can be performed, but the amount of physical memory required increases
Solution Approach 1:
The patent divides the pixel array into multiple smaller blocks, each containing a subset of rows with both active and dark pixels. This segmentation allows the correction algorithm to process one block at a time, collecting statistics only from the dark pixels within the current block. Consequently, only a minimal amount of memory is needed to store the data for a single block during processing, rather than storing multiple entire rows across the full sensor width, significantly reducing the memory capacity requirement.
3Productivity
If localized row-by-row method is used, then processing can be performed, but the time spent processing the data increases
Solution Approach 1:
The patent segments the pixel array into multiple independent blocks that can be processed in parallel. Each block contains sufficient dark pixels for independent offset estimation, allowing the processing pipeline to operate on multiple blocks simultaneously. This parallelization dramatically reduces the total processing time compared to sequential row-by-row processing, as the computational workload is distributed across multiple independent processing units that can execute concurrently.
4Adaptability or versatility
If row-by-row noise correction is applied after interpolation, then correction can be performed on CFA data, but the statistics required to estimate offset are dramatically changed
Solution Approach 1:
The patent performs row noise correction on the raw CFA data before the interpolation (demosaicing) step. By executing the correction in advance, the method operates on the original pixel data with intact color filter array patterns, where dark pixels provide reliable offset statistics. This preliminary action avoids the problem of interpolation altering the data statistics, as the correction is applied to the unprocessed CFA data where the relationship between dark pixels and offset is direct and predictable.
Solution Approach 2:
The patent inverts the conventional processing order by applying row noise correction before interpolation rather than after. This inversion allows the method to exploit the original CFA data structure and dark pixel statistics for accurate offset estimation, avoiding the statistical distortions introduced by interpolation. The inverted sequence maintains the integrity of the dark pixel reference signal throughout the correction process.
Data Source
AI summary
A method for reducing the row noise from complementary metal oxide semiconductor (CMOS) image sensor by using average values from sub-regions of the shielded pixels. The method operates on sensor with and without a Color Filter Array (CFA) before any interpolation is applied and estimates the local offset by subtracting out outliers and averaging the averages of sub-regions in the shielded pixels. The method also reduces the pixel-to-pixel noise while reducing the row noise.