3D OCT Image Denoising via Localized Noise Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for denoising and enhancing 3D OCT images fail to meet real-time processing requirements, inadequately address noise with varying characteristics, and result in unclear details due to limitations of existing filters and processing speeds.
Innovation Solution
Classifying 3D images into sub-areas based on noise characteristics and applying different denoising methods, followed by contrast ratio enhancement through bilinear denoising models and weighted synthesis of high and low frequency components.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If hardware-based denoising with mixed technology (space, frequency, angle, polarization) is used, then noise is reduced, but system manufacturing cost increases and imaging speed decreases
Solution Approach 1:
The patent replaces hardware-based denoising mechanisms with software-based digital filtering methods. Specifically, it uses a mixed model filter that combines linear filtering (for additive noise) and non-linear filtering (for multiplicative noise) in the en face image domain, eliminating the need for additional optical hardware while maintaining denoising effectiveness and preserving imaging speed.
Solution Approach 2:
The patent transforms the denoising approach by changing the domain of processing from time-domain hardware manipulation to frequency-domain digital filtering. It applies Fourier transformation to convert OCT signals into spectral data, then applies filtering operations in the frequency domain, which allows for effective noise reduction without affecting the physical imaging process speed.
2Object-affected harmful factors
If 2D filters are applied to denoise OCT images, then noise is reduced, but processing speed decreases and real-time processing requirement is not met
Solution Approach 1:
The patent segments the 3D OCT dataset into multiple 2D en face images, processes each slice independently with filtering operations, and then reconstructs the 3D volume. This segmentation approach allows for efficient parallel processing of individual slices, reducing overall processing time compared to applying filters to the entire 3D dataset at once, thereby enabling real-time processing.
3Object-affected harmful factors
If filtering is applied to reduce noise, then noise is reduced, but image details become unclear
Solution Approach 1:
The patent applies different filtering strategies to different regions of the image based on local characteristics. It uses a mixed model that adapts the filtering approach: linear filtering for regions dominated by additive noise and non-linear filtering for regions with multiplicative noise. This localized adaptation preserves edge sharpness and fine details while effectively reducing noise in homogeneous regions.
Solution Approach 2:
The patent employs a composite filtering approach by combining multiple filtering techniques (linear and non-linear filters) into a unified mixed model. This composite filter leverages the strengths of each individual filter type: the efficiency of linear filtering for Gaussian noise and the edge-preserving capability of non-linear filtering for speckle noise, achieving both noise reduction and detail preservation.
Data Source
AI summary
A method of enhancing a quality of a 3 dimensional (3D) image includes classifying an input 3D image into a plurality of sub-areas based on noise characteristics of the plurality of sub-areas of the input 3D image, denoising each of the plurality of sub-areas of the input 3D image by using different denoising methods according to noise characteristics of each of the classified plurality of sub-areas and obtaining a second 3D image after the denoising, and enhancing a contrast ratio of the second 3D image after the denoising.