Image Noise Reduction via Spectral Transforms
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image noise reduction techniques are inefficient in detecting and removing row and column noise in imaging systems, particularly in embedded systems with limited computational power, as they require processing of individual pixel values and struggle to distinguish between scene-based information and temporal noise.
Innovation Solution
The use of spectral transforms applied to row and column values to selectively reduce noise by determining spectral coefficients, adjusting them based on noise patterns, and generating correction terms to filter out noise from image frames, thereby reducing spatially correlated temporal noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional pixel-based noise reduction processing is used, then noise detection capability is improved, but computational complexity and memory requirements increase significantly
Solution Approach 1:
The patent segments the image processing task by operating on row and column mean values separately rather than processing all individual pixels. This divides the computational problem into smaller, manageable segments that can be processed independently, reducing overall computational complexity while maintaining noise detection capability.
Solution Approach 2:
The patent extracts only the essential features (row and column mean values) needed for noise detection, rather than processing complete pixel data. This extraction approach retains the ability to detect spatially correlated noise while significantly reducing the data volume requiring computation.
2Measurement precision
If individual pixel processing is performed to identify noise, then noise detection accuracy is improved, but processing time increases
Solution Approach 1:
The patent applies partial action by processing only row and column mean values rather than all pixel data. This partial processing approach is sufficient to detect spatially correlated noise patterns while avoiding the excessive time cost of processing every individual pixel, achieving a balance between detection accuracy and processing speed.
3Productivity
If row and column mean values are processed using spectral transforms, then computational efficiency is improved, but memory requirements are reduced
Solution Approach 1:
The patent changes the parameter representation from individual pixel values to row and column mean values, and further to spectral domain representations. This parameter transformation reduces the quantity of data requiring storage while maintaining the information necessary for noise detection, thereby reducing memory requirements.
Data Source
AI summary
Various techniques are provided for reducing noise in captured image frames. In one example, a method includes determining row values for image frames comprising scene information and noise information. The method also includes performing first spectral transforms in a first domain on corresponding subsets of the row values to determine first spectral coefficients. The method also includes performing second spectral transforms in a second domain on corresponding subsets of the first spectral coefficients to determine second spectral coefficients. The method also includes selectively adjusting the second spectral coefficients. The method also includes determining row correction terms based on the adjusted second spectral coefficients to reduce the noise information of the image frames. Additional methods and systems are also provided.