3D Fourier Transform Denoising for Image Sequences
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional noise reduction methods for images and image sequences, particularly in fields like medical imaging and astrophotography, face limitations due to the need for motion estimation, which is ill-posed and ineffective for non-temporal data sets, and often compromise dynamic range and resolution.
Innovation Solution
An adaptive filtering method in the Fourier domain that treats 3-D images as 'chunks' across all three dimensions, exploiting coherence between time and space to distinguish between image signals and noise without requiring motion estimation, using techniques like chunking, apodization, and Wiener filtering to preserve image features while attenuating noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If conventional denoising methods use Fourier transforms with motion estimation, then noise reduction is achieved, but the method becomes ill-posed and limits applicability to non-temporal data sets
Solution Approach 1:
The patent extends conventional 2-D Fourier transform denoising to 3-D Fourier transform by adding the temporal dimension as a third dimension. This allows the method to process image sequences and non-temporal data sets uniformly, resolving the limitation of motion estimation methods that cannot handle non-temporal data. The 3-D Fourier transform treats spatial and temporal frequencies symmetrically, enabling versatile application across different data types.
Solution Approach 2:
The patent extracts and removes the motion estimation step from the denoising process. By directly applying 3-D Fourier transform to the image sequence without motion compensation, the method eliminates the ill-posed nature of motion estimation while preserving the ability to exploit temporal redundancy for noise reduction.
2Ease of manufacture
If conventional denoising methods use 2-D transforms followed by 1-D transforms, then processing is simplified, but noise reduction effectiveness is limited compared to true 3-D denoising
Solution Approach 1:
The patent merges the 2-D spatial transform and 1-D temporal transform into a unified 3-D Fourier transform operation. This combination allows simultaneous exploitation of spatial and temporal correlations in a single processing step, achieving superior noise reduction while maintaining computational efficiency through the separable property of the 3-D transform.
3Object-affected harmful factors
If noise reduction techniques are applied to image sequences, then dynamic range is improved, but spatial and temporal resolution may be compromised
Solution Approach 1:
The patent applies local quality by performing denoising in the frequency domain where different frequency components can be treated differently. By operating in the Fourier domain, the method can selectively reduce noise in high-frequency regions while preserving low-frequency signal components, thereby improving dynamic range without compromising spatial and temporal resolution.
Data Source
AI summary
A method of mitigating noise in source image data representing pixels of a 3-D image. The “3-D image” may be any type of 3-D image, regardless of whether the third dimension is spatial, temporal, or some other parameter. The 3-D image is divided into three-dimensional chunks of pixels. These chunks are apodized and a three-dimensional Fourier transform is performed on each chunk, thereby producing a three-dimensional spectrum of each chunk. The transformed chunks are processed to estimate a noise floor based on spectral values of the pixels within each chunk. A noise threshold is then determined, and the spectrum of each chunk is filtered with a denoising filter based on the noise threshold. The chunks are then inverse transformed, and recombined into a denoised 3-D image.