Multi-resolution Image Noise Removal via Band-limited Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods struggle to accurately remove noise while preserving image structure, as they often require a trade-off between noise removal and retention of effective image structure information, and lack flexibility in adjusting frequency characteristics of noise and edge components.
Innovation Solution
An image processing method that generates multiple-resolution images, extracts edge components using band pass filters, modulates weights for edge components across frequency bands, and synthesizes these components to emphasize edges or remove noise effectively, allowing for flexible noise extraction and preservation of image structure.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If noise removal processing is applied to subband coefficients from multiple-resolution transformation, then noise components are removed, but image structure information is destroyed
Solution Approach 1:
The image is decomposed into multiple resolution levels through multi-scale decomposition, separating noise and edge components at different frequency bands. This segmentation allows selective processing of noise at each scale without affecting the overall image structure, as each resolution level can be independently processed and then reconstructed.
Solution Approach 2:
Different processing strategies are applied to different frequency subbands based on their local characteristics. High-frequency subbands undergo aggressive noise removal while low-frequency subbands preserve more structure information. The processing intensity is locally adapted to the noise and signal characteristics of each subband, enabling effective noise removal without uniform destruction of image structure.
2Object-affected harmful factors
If aggressive noise removal is applied to completely remove noise components, then noise is eliminated, but image structure is destroyed
Solution Approach 1:
The image undergoes preliminary multi-scale decomposition before noise removal,预先 separating noise and structure components across different resolution levels. This preliminary action allows the subsequent noise removal to target only noise components at each scale, preventing the destruction of image structure that would occur with direct aggressive filtering on the original image.
Solution Approach 2:
The noise removal process dynamically adjusts processing intensity based on the local noise characteristics and signal strength at each resolution level. Rather than applying a fixed aggressive filter, the method adapts the processing strength to preserve image structure while removing noise, with stronger processing applied where noise dominates and weaker processing where structure is critical.
3Ease of operation
If a single noise removal rate is applied to the entire image, then processing is simple, but residual noise and structure loss cannot be optimized
Solution Approach 1:
The single noise removal rate is segmented into multiple removal rates applied at different resolution levels. Each scale receives a customized noise removal rate based on its specific noise characteristics, allowing optimization of noise removal accuracy without requiring complex manual tuning of a single global parameter. The segmentation transforms a simple but ineffective single-rate approach into a multi-rate system that maintains operational simplicity while improving precision.
Solution Approach 2:
The noise removal rate parameter is changed across different resolution levels rather than remaining constant. This parameter variation allows the system to optimize noise removal at each scale according to the local noise-to-signal ratio, achieving higher overall noise removal accuracy while maintaining a relatively simple processing framework that automatically adapts the parameter based on the decomposition level.
4Productivity
If only high-frequency subband coefficients are processed for noise removal, then processing is efficient, but low-frequency noise components remain
Solution Approach 1:
The noise removal process is segmented across multiple frequency subbands including both high-frequency and low-frequency components. Rather than processing only high-frequency subbands, the method applies noise removal to each subband according to its characteristics, ensuring that low-frequency noise components are also addressed. This segmentation maintains efficiency by processing each subband independently while comprehensively removing noise across the full frequency spectrum.
Solution Approach 2:
The method applies partial noise removal action to each subband based on its specific noise content rather than uniformly processing all subbands at maximum intensity. Low-frequency subbands receive appropriate noise removal action tailored to their noise characteristics, preventing residual low-frequency noise while avoiding excessive processing that would destroy structure. This partial action approach optimizes the balance between noise removal completeness and processing efficiency.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An image processing method adopted to remove noise present in an image, includes: an image input step in which an original image constitutedof apluralityof pixels is input; a multiple-resolution images generation step in which a plurality of band-limited images with resolutions decreasing in sequence are generated by filtering the input original image; a first noise removal step in which virtual noise removal processing is executed individually for each of the band-limited images; a second noise removal step in which actual noise removal processing is executed for the individual band-limited images based upon the band-limited images from which noise has been virtually removed through the first noise removal step; and an image acquisition step in which a noise-free image of the original image is obtained based upon the individual band-limited images from which noise has been actually removed through the second noise removal step, and the virtual noise removal processing executed in the first noise removal step and the actual noise removal processing executed in the second noise removal step are differentiated in correspondence to a frequency band of a band-limited image.