Defective Pixel Fixing Using Gradient Analysis and Median Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image sensors face challenges with defective pixels, which can be transient and not accurately identified by pre-provided defective pixel maps, leading to inaccurate correction and the need for methods to determine and correct defective pixels without relying on such maps.
Innovation Solution
A method and unit for defective pixel fixing that determine candidate values for a target pixel by analyzing gradients in multiple directions and using these values to predict and replace the pixel values, with a median-based approach to reduce outlier impact, allowing for selective replacement of defective pixel values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If pre-provided defective pixel maps are used to identify defective pixels, then the identification process is simplified, but the accuracy deteriorates because transient defects are not captured and defective pixels may change over time
Solution Approach 1:
The patent transitions from static defective pixel maps to dynamic real-time detection by analyzing gradient changes across multiple frames. The system continuously updates defective pixel identification based on current image data, allowing it to adapt to transient defects that change over time. This is achieved by calculating gradients between current and reference frames and detecting abnormal gradient patterns that indicate defective pixels.
Solution Approach 2:
The system implements feedback by using detected defective pixel information to improve subsequent detection accuracy. The gradient analysis results from multiple frames are combined and used to refine the identification of defective pixels. The system continuously monitors and adjusts its defective pixel map based on ongoing gradient analysis, creating a closed-loop feedback mechanism that improves accuracy over time.
2Device complexity
If traditional averaging methods are used to predict defective pixel values, then the correction process is simple, but the prediction accuracy deteriorates due to outlier impact
Solution Approach 1:
The patent extracts and removes outlier values from the gradient analysis before calculating the predicted pixel value. By identifying and excluding extreme gradient values that do not represent the true local pattern, the system prevents these outliers from skewing the average. This extraction of harmful elements (outliers) allows the remaining values to provide a more accurate basis for prediction.
Solution Approach 2:
The system changes the parameter used for prediction from a simple average to a robust statistical measure that accounts for gradient distribution. Instead of directly averaging pixel values, the patent analyzes gradient patterns, determines candidate values based on gradient magnitude and direction, and selects the most representative value. This parameter transformation from direct averaging to gradient-based candidate selection improves accuracy while managing complexity.
Data Source
AI summary
Methods of fixing defective pixels are described wherein a predicted value for a target pixel in a target color channel is determined based on the values of nearby pixels, wherein the target pixel value can be selectively replaced with the predicted value. The predicted value is determined by determining a candidate value for each of a plurality of directions using: (i) a gradient of pixel values in one color channel along the respective direction and (ii) a pixel value of a pixel in the target color channel which is aligned with the target pixel along the respective direction. Using gradients can provide better predicted values than averaging nearby pixel values since rates of change of pixel values are taken into account. The median of the candidate values may be used in order to reduce the impact of other defective pixels on the predicted value for the target pixel.