Signal Processing Apparatus for Image Sensor Noise Correction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image capturing devices face challenges in accurately correcting dark current and noise from image sensors, particularly due to variations in circuit factors and manufacturing defects, which affect image quality.
Innovation Solution
A signal processing apparatus utilizing a machine learning model to generate control data for correcting image data by differentiating between light-receiving and light-blocking regions, applying this control data to reduce fixed pattern noise and shading components through OB clamp processing, and adapting to variations in image capturing conditions and chip-specific noise patterns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a trained model generated through machine learning is used to correct image data, then the adaptability to various shooting conditions is improved, but the device complexity increases
Solution Approach 1:
The patent divides the image sensor output into multiple regions: light-receiving regions (containing actual image data) and light-blocking regions (containing reference data for correction). This segmentation allows the system to process different types of data through different pathways, reducing the complexity of applying machine learning to the entire image while maintaining adaptability.
Solution Approach 2:
The patent introduces control data as an intermediary between the machine learning model and the correction process. The control data generation unit uses the trained model to generate control data, which is then used by the signal processing unit to perform correction. This intermediary approach separates the complex machine learning inference from the actual correction operation, reducing device complexity.
2Measurement precision
If correction values are pre-stored for multiple division patterns, then the correction accuracy for various shooting conditions is improved, but the manufacturing complexity increases
Solution Approach 1:
The patent implements dynamic block division where the image capturing region is divided into multiple blocks with variable sizes and positions based on shooting conditions. This dynamic approach allows the system to adapt to different scenarios without requiring pre-stored correction values for every possible condition, reducing manufacturing complexity while maintaining correction accuracy.
Solution Approach 2:
The patent changes the parameters of block division (number of blocks, block size, block positions) based on shooting conditions. By adjusting these parameters dynamically, the system can achieve accurate correction for various conditions without manufacturing multiple fixed correction value sets, thereby reducing manufacturing complexity.
3Measurement precision
If the trained model is applied to correct all image data, then the noise reduction accuracy is improved, but the processing time increases
Solution Approach 1:
The patent applies the trained model only to generate control data from light-blocking regions, rather than applying it to correct all image data. The signal processing unit then uses this control data to correct the light-receiving regions. This partial application of the model reduces processing time while maintaining noise reduction accuracy through the correction process.
Data Source
AI summary
A signal processing apparatus that processes image data output from a photoelectric conversion unit including a light-receiving region and a light-blocking region. The apparatus includes a control data generation unit that outputs control data used to generate correction data for correcting the image data using a trained model generated through machine learning, and a signal processing unit that generates the correction data on the basis of light-blocked image data and the control data, the light-blocked image data being image data, among the image data, that is from the light-blocking region, and corrects light-received image data in accordance with the correction data without applying the trained model, the light-received image data being image data, among the image data, that is from the light-receiving region.