Multi-branch Neural Network for Artifact-Free Image Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image segmentation systems using neural networks require significant computing resources and processing time, often producing inaccurate segmentation masks with boundary artifacts due to the inclusion of post-processing stages.
Innovation Solution
A multi-branch neural network architecture that combines high-level and low-level feature maps from separate encoder branches, processed by a boundary refinement module, to generate accurate segmentation data without the need for post-processing stages, reducing resource consumption and processing time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If post-processing stages are included in the image segmentation system, then segmentation accuracy is improved, but processing speed decreases and computing resources increase
Solution Approach 1:
The patent combines multiple encoder branches (high-level and low-level) into a unified neural network architecture that directly outputs segmentation masks without requiring separate post-processing stages. The boundary refinement module integrates boundary enhancement functions within the main network, eliminating the need for external post-processing while maintaining accuracy and improving speed.
2Measurement precision
If post-processing stages are included in the image segmentation system, then segmentation accuracy is improved, but computing system resources increase
Solution Approach 1:
The patent merges boundary refinement functionality into the main neural network architecture through a dedicated module that processes features during the forward pass. This integration eliminates the need for separate post-processing systems, reducing overall computing resource requirements while maintaining segmentation accuracy.
3Device complexity
If a single encoder neural network is used, then device complexity is reduced, but segmentation accuracy decreases due to boundary artifacts
Solution Approach 1:
The patent divides the encoding process into separate high-level and low-level encoder branches, each specialized for different feature extraction tasks. The low-level encoder captures detailed boundary information while the high-level encoder captures semantic information, and their features are combined to produce accurate segmentation masks without boundary artifacts.
Data Source
AI summary
A multi-branch neural network generates segmentation data for a received image. The received image is provided to a high-level branch and a low-level branch. Based on the received image, the high-level branch generates a feature map of high-level image features, and the low-level branch generates a feature map of low-level image features. The high-level feature map and the low-level feature map are combined to generate a combined feature map. The combined feature map is provided to a boundary refinement module that includes a dense-connection neural network, which generates segmentation data for the received image, based on the combined feature map.