Image Sensor Dithering for Sub-Code Fixed Pattern Noise Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image processing techniques are limited in compensating for row or column noise in image sensors due to the inability to add or subtract less than one code, which is costly to improve with higher accuracy ADCs, leading to suboptimal image quality.
Innovation Solution
A method of image processing that involves reading out image data from a pixel array and adding a noise contribution derived from correction data, where different correction data are used for different groups of pixels associated with specific pixel value DC shifts, using a combination of M-bit reference words to derive a statistically weighted pseudo-random dither, which can be applied differently to groups of pixels sharing common design factors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If higher resolution ADCs are used to reduce quantization error, then measurement precision is improved, but device complexity and cost increase
Solution Approach 1:
The patent introduces correction data as an intermediary element that mediates between the limited-resolution ADC output and the desired high-precision image data. The correction data, stored in a correction buffer, serves as a compensating mechanism that adds fine-grained adjustment values to the quantized pixel data, effectively bridging the precision gap without requiring a higher-resolution ADC
Solution Approach 2:
The patent changes the parameter of data representation by introducing correction data with higher precision than the original ADC output. Instead of changing the ADC hardware resolution, the system changes the effective precision parameter through software-based correction values that are added to the quantized pixel data, transforming the output precision without transforming the hardware capability
2Measurement precision
If dithering is applied to reduce quantization artifacts, then image quality is improved, but noise is introduced to the image data
Solution Approach 1:
The patent converts the harmful quantization error into a beneficial correction opportunity. By analyzing the quantization error pattern and generating correction data that specifically addresses these errors, the system transforms what was previously a harmful artifact into a target for improvement. The correction data effectively reverses the harmful effect of quantization without introducing additional noise
Solution Approach 2:
Instead of adding random noise to mask quantization artifacts (conventional dithering), the patent inverts the approach by subtracting calculated correction values from the quantization error. The correction data is derived by inverting the quantization process mathematically, allowing the system to remove artifacts rather than mask them, thereby improving image quality without introducing the noise inherent in traditional dithering methods
Data Source
AI summary
An image processor includes a readout arranged to read out an M-bit image data word from an image sensor pixel array and an adder arranged to add a noise contribution to the image data word to obtain a dithered M-bit word. A dither processor is arranged to derive correction data having a word size of M+1 bits from a combination of a plurality of M-bit reference words. The noise contribution are derived from said correction data, wherein different correction data are derived for different groups of pixels, each different group of pixels is associated with a specific pixel value DC shift.