Image Processor Noise Reduction via Autocorrelation and Median Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods fail to effectively detect and reduce random noise in image data from solid-state imaging devices, particularly under low luminance conditions, as random noise lacks periodicity and is difficult to suppress using existing autocorrelation techniques.
Innovation Solution
A method involving an image processor that extracts pixel values from image data, calculates autocorrelation coefficients, and compares them against threshold values to identify and separate pixels affected by random noise, allowing for noise reduction and correction, including camera shake adjustment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If autocorrelation method is used to reduce noise, then fixed pattern noise can be suppressed, but random noise cannot be effectively reduced because it lacks periodicity
Solution Approach 1:
The patent segments the noise reduction process into two distinct stages: first applying autocorrelation method to suppress fixed pattern noise, then applying median filtering to address random noise. This segmentation allows each method to be optimized for its specific noise type, resolving the contradiction between effectiveness for fixed pattern noise and adaptability to random noise.
Solution Approach 2:
The patent merges two different noise reduction techniques (autocorrelation method and median filtering) into a single processing pipeline. The autocorrelation method handles fixed pattern noise while the median filtering handles random noise, combining their strengths to achieve comprehensive noise suppression that neither method could achieve alone.
2Reliability
If median filtering is applied to reduce random noise, then noise levels can be reduced, but computation time increases
Solution Approach 1:
The patent applies preliminary action by first using the autocorrelation method to suppress fixed pattern noise before applying median filtering. This preliminary processing reduces the overall noise burden, allowing the median filtering step to focus only on random noise, thereby reducing the computation time required for this specific operation.
Solution Approach 2:
By segmenting the noise reduction into two stages with specific focuses, the patent avoids the need for extensive median filtering across all pixel data. The first stage handles deterministic noise while the second stage addresses stochastic noise, reducing the total computational burden compared to applying only median filtering.
3Manufacturing precision
If image processing is performed under low luminance conditions, then image quality can be maintained, but noise levels increase
Solution Approach 1:
The patent converts the harmful effect of increased noise under low luminance conditions into a beneficial process by using the noise characteristics themselves as guidance for filtering. The autocorrelation method exploits the periodicity of fixed pattern noise to identify and remove it, while the median filtering uses noise statistical properties to distinguish signal from noise, thereby improving image quality despite low light conditions.
Data Source
AI summary
A method of processing noise in image data by an image processor having a signal-processing portion converting an image signal from an image sensor into a digital signal and outputting the converted signal as image data for each frame, the image data indicating sets of pixel values each having a brightness at a corresponding one of coordinate points arranged in directions of rows and columns is disclosed. The method includes the steps of: extracting pixel values; deciding pixel value; finding autocorrelation coefficients of pixel values which are less than a first threshold value; and deciding random noise in the image.