Dual Encoding-Decoding Neural Network for Image Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image processing methods using deep learning technologies face challenges in achieving robust and precise image segmentation, particularly in varying light environments and across different imaging devices, with existing CNN models requiring extensive manual operations and lacking stability in segmentation results.
Innovation Solution
The proposed method employs a neural network architecture comprising two encoding-decoding networks with concatenation layers, utilizing encoding and decoding meta-networks with down-sampling and up-sampling layers, and residual modules to segment images into distinct regions, improving segmentation accuracy and reducing manual intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If a single encoding-decoding network is used for image segmentation, then the device complexity is low, but the segmentation precision and robustness are insufficient
Solution Approach 1:
The patent divides the image segmentation task into two sequential stages using two separate encoding-decoding networks. The first network performs initial segmentation to obtain a first segmented image and first output feature map, while the second network performs refined segmentation to obtain a second segmented image. This segmentation of the segmentation task allows each network to focus on specific aspects, improving overall precision without requiring a single overly complex network
Solution Approach 2:
The patent implements a nested architecture where the output of the first encoding-decoding network serves as input to the second encoding-decoding network. Specifically, the first segmented image and first output feature map are fed into the second network for further processing. This nested structure allows the system to build upon previous segmentation results, enhancing precision through progressive refinement while maintaining manageable complexity at each stage
2Reliability
If extensive manual operations are performed for image segmentation, then the segmentation results can be adjusted, but the productivity and automation level are reduced
Solution Approach 1:
The patent implements an end-to-end automated segmentation system where the dual encoding-decoding networks perform segmentation tasks autonomously without requiring manual intervention. The networks self-adjust through their internal feature extraction and refinement mechanisms, with the second network automatically refining the output of the first network. This self-service capability maintains segmentation stability through consistent automated processing while maximizing productivity by eliminating manual operations
Solution Approach 2:
The patent incorporates feedback mechanisms where the output of each network stage is fed back into the next stage for further processing. The first segmented image and first output feature map serve as feedback input to the second network, allowing automatic refinement and correction. This feedback loop ensures segmentation stability through iterative improvement while maintaining full automation, as the system self-corrects without human intervention
3Adaptability or versatility
If a simple encoding-decoding network is used, then the ease of operation is high, but the segmentation accuracy varies across different imaging conditions
Solution Approach 1:
The patent designs two encoding-decoding networks with universal architectures that can handle various imaging conditions and devices. Each network is capable of performing both encoding (feature extraction) and decoding (segmentation) functions, and the combined system can adapt to different input types including images from various light environments and imaging devices. This multi-functionality achieves high generalization capability while keeping each individual network relatively simple
Solution Approach 2:
The patent adds a temporal or sequential dimension to the segmentation process by using two networks in sequence rather than one network. The first network processes the input image to produce initial segmentation, and the second network processes this output to produce refined segmentation. This dimensional expansion from single-stage to multi-stage processing enhances adaptability across different imaging conditions without requiring each individual network to be overly complex
Data Source
AI summary
Disclosed are an image processing method, an image processing device, a neutral network and a training method thereof, and a storage medium. The image processing method includes: obtaining an input image; performing a segmentation process on the input image via a first encoding-decoding network, to obtain a first output feature map and the first segmented image; concatenating the first output feature map with at least one selected from the group consisting of the input image and the first segmented image, to obtain an input of the second encoding-decoding network; and performing a segmentation process on the input of the second encoding-decoding network via a second encoding-decoding network, to obtain the second segmented image. And the first encoding-decoding network and the second encoding-decoding network forms a neural network.