Image Noise Reduction via Wavelet Coefficient Energy Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing noise reduction methods for images captured at high ISO settings are inadequate, as they often assume Gaussian white noise, leading to either insufficient noise removal or loss of image detail, and are computationally expensive.
Innovation Solution
A method using wavelet decomposition to generate coefficients at different scales, modifying them based on normalized inter-scale energy, single scale energy, and weighted average single scale energy, and reconstructing the image to effectively suppress noise while preserving image details.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If wavelet thresholding methods are used to remove noise, then noise reduction is achieved, but image detail may be lost or insufficient noise removal occurs due to arbitrary threshold selection
Solution Approach 1:
The patent changes the parameter selection approach from arbitrary fixed thresholds to data-driven thresholds based on wavelet coefficient statistical analysis. By calculating the standard deviation of wavelet coefficients and using it to determine adaptive threshold values, the method dynamically adjusts parameters according to the actual noise characteristics in the image, thereby resolving the contradiction between noise removal and detail preservation.
Solution Approach 2:
The patent implements feedback by using the calculated wavelet coefficient statistics (mean and standard deviation) to inform the threshold selection process. The threshold is determined based on feedback from the actual coefficient distribution in the image, allowing the method to adapt to different noise levels and image characteristics, thus preventing both over-thresholding (detail loss) and under-thresholding (insufficient noise removal).
2Object-affected harmful factors
If sophisticated noise reducing algorithms are used, then noise removal effectiveness is improved, but processing time increases significantly
Solution Approach 1:
The patent applies segmentation by dividing the image into different frequency subbands through wavelet decomposition. This allows noise reduction to be applied selectively to different frequency components, processing only the relevant high-frequency noise-containing coefficients while preserving low-frequency image structure. This segmented approach reduces computational complexity compared to processing the entire image uniformly with sophisticated algorithms.
Solution Approach 2:
The patent extracts the essential noise characteristics by calculating only the mean and standard deviation of wavelet coefficients, then uses these extracted parameters to determine thresholds for noise removal. By taking out only the necessary statistical information rather than applying complex full-image analysis, the method achieves effective noise reduction with minimal processing time.
3Device complexity
If anisotropic diffusion methods are used assuming white noise, then processing is simplified, but noise removal is ineffective for high ISO camera noise
Solution Approach 1:
The patent changes the noise model parameter from the assumed white noise model to a model based on actual high ISO camera noise characteristics. By calculating wavelet coefficient statistics from the image data itself and using these to determine adaptive thresholds, the method adapts to the specific noise characteristics of high ISO images without requiring complex noise modeling, thus maintaining simplicity while improving effectiveness.
Data Source
AI summary
A method of reducing noise in an image comprises decomposing the image to generate wavelet coefficients at different scales. The wavelet coefficients are then modified based on the energy of the wavelet coefficients at the different scales. The image is reconstructed based on the modified wavelet coefficients.