ES-Hybrid Filtering for Edge-Preserving Thermal Image Denoising
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing noise removal methods in real-time systems, particularly in thermal images, cause structural distortions and edge loss, leading to excessive DSP consumption, high memory usage, and inconsistent results across different noise types.
Innovation Solution
A hybrid method using median filtering and convolutional filtering, combined with X-Model and Plus-Model filters, is applied to preserve edges and structure in thermal images, optimizing image quality without significant computational or memory resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If conventional noise removal methods are applied, then noise is reduced, but structural distortions and edge loss occur
Solution Approach 1:
The patent divides the noise removal process into multiple stages: first applying a median filter to remove impulse noise while preserving edges, then applying a convolutional filter to remove Gaussian noise. This segmented approach allows each filter to target specific noise types without compromising image structure, resolving the contradiction between noise reduction and structural preservation.
Solution Approach 2:
The patent employs different filtering strategies for different regions and noise types within the image. The median filter is applied primarily for edge preservation in regions with impulse noise, while the convolutional filter handles Gaussian noise in other regions. This localized application of different filtering qualities ensures structural integrity is maintained while effectively removing various noise types.
2Manufacturing precision
If advanced filtering methods are used to preserve edges, then edge preservation improves, but DSP consumption increases
Solution Approach 1:
The patent combines median filtering and convolutional filtering into a single hybrid processing pipeline. By merging these two filtering approaches, the system achieves superior edge preservation and noise removal performance while optimizing the overall computational efficiency. The combined approach allows shared resource utilization and coordinated processing that reduces total DSP consumption compared to applying filters separately.
Solution Approach 2:
The patent implements adaptive filtering where the processing intensity and filter parameters are dynamically adjusted based on local image characteristics and detected noise types. This dynamic approach allows the system to apply stronger filtering only where necessary, reducing overall DSP consumption while maintaining high edge preservation quality in critical regions.
3Object-affected harmful factors
If multiple filtering operations are applied, then noise removal effectiveness improves, but memory usage increases
Solution Approach 1:
The patent performs preliminary noise classification and region identification before applying the full filtering sequence. By preliminarily analyzing the image to identify noise types and affected regions, the system can apply filtering operations only where needed, reducing memory requirements for storing intermediate results and processing the entire image uniformly.
Solution Approach 2:
The patent implements an efficient memory management strategy where intermediate filtering results are discarded after being used for the next processing stage, rather than being stored in memory. The median filter output is immediately fed into the convolutional filter without intermediate storage, and final results are produced in a streaming manner, significantly reducing peak memory usage while maintaining noise removal effectiveness.
4Productivity
If real-time processing is implemented, then processing speed improves, but noise removal quality may deteriorate
Solution Approach 1:
The patent segments the noise removal task into two parallel processing streams: one for impulse noise using median filtering and another for Gaussian noise using convolutional filtering. This segmentation allows independent optimization of each processing path for real-time performance while maintaining comprehensive noise removal quality. Each segment can be processed at optimal speed without waiting for the other, improving overall processing throughput.
Solution Approach 2:
The patent merges the outputs of the median filter and convolutional filter through a sophisticated combination strategy that produces the final denoised image. This merging operation is designed to be computationally efficient, using simple pixel-wise operations that maintain real-time processing speed while preserving the noise removal quality benefits of both filtering approaches.
Data Source
Figure 1~2
Figure 3~5
Figure 6~8
AI summary
The invention relates to the ES-Hybrid method in daytime/thermal images that provides noise removal by preserving edges and structure in which sorting-based median filtering and convolutional filters are used.