Modular Neural Network Building Units for Image Denoising
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High-resolution imaging techniques, such as FIB-SEM, produce noisy images due to optical and thermal effects, making it difficult to obtain noise-free images, especially at nanometer scales, and existing machine learning models require ground truth images for training, which are often unavailable.
Innovation Solution
A machine learning model that uses sequences of images to identify and remove noise without ground truth images, employing a modular architecture with neural networks and noise attention blocks to denoise images by comparing images in sequences, allowing for training with noisy image pairs and iterative noise removal processes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-resolution imaging techniques are used, then image resolution is improved, but noise increases
Solution Approach 1:
The patent segments the image denoising task into multiple processing stages using sequential building units. Each building unit processes a portion of the input image and passes results to the next unit, enabling progressive denoising while preserving high-resolution details through staged processing rather than single-step operation
Solution Approach 2:
The patent introduces an intermediary loss function that guides the denoising process by comparing intermediate processing results with reference images. This intermediary evaluation mechanism allows the system to progressively remove noise while maintaining resolution without requiring direct ground truth comparison at each stage
2Measurement precision
If ground truth images are used for training machine learning models, then model accuracy is improved, but data availability decreases
Solution Approach 1:
The patent inverts the traditional training approach by using noisy images as reference data instead of requiring clean ground truth images. The building units are trained to transform noisy images into denoised versions, eliminating the dependency on unavailable ground truth data while maintaining model accuracy through this reversed training paradigm
Solution Approach 2:
The system performs self-service training by using its own processed outputs as reference data for subsequent training iterations. The model processes noisy images, uses the results as new reference points, and continues training without external ground truth data, enabling continuous improvement with available noisy image pairs
3Measurement precision
If iterative noise removal processes are used, then image quality is improved, but processing time increases
Solution Approach 1:
The patent segments the iterative denoising process into parallel building units that can process different regions or aspects of the image simultaneously. This segmentation enables iterative refinement without sequential processing delays, maintaining image quality improvement while reducing overall processing time through parallel execution
Solution Approach 2:
The patent applies partial denoising actions in each building unit stage rather than attempting complete denoising in a single pass. Each unit performs a portion of the noise removal task, and multiple units work together to achieve comprehensive denoising, balancing processing speed with image quality through distributed partial actions
Data Source
AI summary
In some examples, a machine learning model may be trained to denoise an image. In some examples, the machine learning model may identify noise in an image of a sequence based at least in part, on at least one other image of the sequence. In some examples, the machine learning model may include a recurrent neural network. In some examples, the machine learning model may have a modular architecture including one or more building units. In some examples, the machine learning model may have a multi-branch architecture. In some examples, the noise may be identified and removed from the image by an iterative process.