Image Denoiser Using Wavelet Scale Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image denoising techniques fail to effectively capture the piecewise smooth nature of objects, leading to issues like noise masking, artificial sharpening, and over-blurring, particularly in color images, and often introduce artifacts such as ringing or over-smoothing.
Innovation Solution
The Image Denoiser decomposes images into piecewise smooth regions, estimates statistical models for each region, and applies local blur estimation to preserve natural edges, using a probabilistic model to generate a denoised image that approximates the original blur, thereby reducing artificial sharpening and maintaining image softness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If conventional denoising schemes use wavelet-based coring techniques, then noise suppression is improved, but ringing artifacts are introduced
Solution Approach 1:
The image is divided into multiple scale levels through wavelet decomposition, with different denoising strategies applied to different scales. Fine scales preserve high-frequency details while coarse scales remove noise, preventing ringing artifacts that occur when uniform coring is applied across all scales.
Solution Approach 2:
The denoising operation adapts local quality parameters based on image content. The coring function strength varies spatially and across scales, being more aggressive in noise-dominated regions and gentler in detail-rich regions, thus suppressing noise while preserving edges and avoiding ringing.
2Object-affected harmful factors
If Gaussian filtering is used to denoise images, then noise suppression is improved, but edges become blurred
Solution Approach 1:
The filtering operation is segmented across multiple wavelet scales, allowing different levels of smoothing at different frequencies. Coarse scales receive stronger Gaussian filtering for noise suppression, while fine scales maintain minimal filtering to preserve edge sharpness.
Solution Approach 2:
The filtering strength dynamically adapts based on local image characteristics. The coring parameter adjusts automatically according to the estimated noise level and local variance, applying stronger filtering where needed and weaker filtering where edges exist, thus achieving both noise suppression and edge preservation.
3Manufacturing precision
If anisotropic diffusion is used to preserve edges, then edge sharpness is improved, but artificial sharpening and cartoonish appearance occur
Solution Approach 1:
Instead of applying diffusion to sharpen edges, the patent uses wavelet coring to selectively remove noise while preserving the natural softness of edges. The approach inverts the conventional wisdom by accepting some edge softening as desirable to avoid the harmful artificial sharpening effect.
Solution Approach 2:
The coring parameter is carefully controlled and adjusted based on local image statistics. By changing the coring strength parameter dynamically across different regions and scales, the method achieves natural-looking denoising without the cartoonish appearance that results from fixed-parameter anisotropic diffusion.
4Object-affected harmful factors
If bilateral filtering is used to preserve edges, then noise suppression is improved, but over-smoothing and over-sharpening occur
Solution Approach 1:
The denoising is segmented across wavelet scales rather than applying a single bilateral filter across the entire image. This multi-scale approach allows noise suppression at coarse scales while preserving natural edge transitions at fine scales, avoiding the over-smoothing and over-sharpening artifacts of conventional bilateral filtering.
Solution Approach 2:
The patent replaces the mechanical bilateral filtering mechanism with a statistical wavelet-based coring approach. Instead of relying on spatial and range weighting that causes artifacts, the method uses probabilistic modeling of wavelet coefficients to achieve noise suppression while maintaining natural image appearance.
Data Source
AI summary
An “Image Denoiser” provides a probabilistic process for denoising color images by segmenting an input image into regions, estimating statistics within each region, and then estimating a clean (or denoised) image using a probabilistic model of image formation. In one embodiment, estimated blur between each region is used to reduce artificial sharpening of region boundaries resulting from denoising the input image. In further embodiments, the estimated blur is used for additional purposes, including sharpening edges between one or more regions, and selectively blurring or sharpening one or more specific regions of the image (i.e., “selective focus”) while maintaining the original blurring between the various regions.