CMOS Image Sensor Noise Reduction via Weighted Median Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
CMOS image sensors face noise issues due to small MOS transistors, particularly random telegraph signal (RTS) noise, which persists despite advanced fabrication techniques, affecting image quality and processing speed.
Innovation Solution
Implementing a method that involves generating multiple samples of analog image data during correlated double sampling periods, applying weighting factors to samples based on their timing, and using median filtering to reduce noise by replacing outlier samples with thresholds, thereby filtering out short-duration RTS events.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Volume of moving object
If smaller MOS transistors are used to reduce image sensor physical size, then miniaturization is achieved, but noise (particularly RTS noise) increases
Solution Approach 1:
The patent segments the sampling process into multiple discrete samples taken at different time points during the integration period. By dividing the single measurement into multiple temporal segments, the system can identify and eliminate noise events that occur only during specific time windows, thereby reducing RTS noise while maintaining small transistor dimensions
Solution Approach 2:
The patent performs preliminary noise filtering by taking multiple samples before final image processing. By preemptively capturing multiple time-stamped samples and identifying outliers during the sampling phase, the system prepares clean data for subsequent processing, reducing the impact of RTS noise before it affects the final image quality
2Object-generated harmful factors
If multiple samples are taken during correlated double sampling periods, then noise reduction is achieved, but processing time increases
Solution Approach 1:
The patent replaces complex temporal noise analysis with a simplified statistical approach. Instead of analyzing the full time-dependent behavior of noise signals, the system uses median filtering and threshold-based outlier rejection on multiple samples, substituting mechanical/temporal processing with statistical computation that achieves noise reduction more efficiently
Solution Approach 2:
The patent changes the parameter being measured from continuous analog signals to discrete digital sample values. By converting analog image data to digital samples during the sampling process, the system enables efficient computational noise filtering using median and mean calculations, reducing processing time compared to analog noise filtering methods
Data Source
AI summary
A method for preprocessing analog image data to reduce noise in the analog image data that is readout from a pixel array of an image sensor during a sampling time is disclosed. The method includes generating multiple samples of the analog image data during the sampling time and then limiting values of the multiple samples to an upper and lower threshold. The method also includes pre-conditioning the multiple samples by applying a weighting factor to each of the multiple samples in response to when a respective sample was generated during the sampling time. A median value of the multiple samples is then determined and outputted.