Image Segmentation Learning with Geometric Edge Parameters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image segmentation techniques, such as those described in Patent Literature 1, suffer from low estimation accuracy, particularly in the distortion of edges, which are often estimated in a zigzag shape when they should be linear or geometrically accurate.
Innovation Solution
A learning device that utilizes a neural network to input learning images with correct edge and geometric parameter data, calculates a cost for evaluation, and updates model parameters to improve edge estimation accuracy by incorporating geometric pattern information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional neural network-based image segmentation is used, then general edge detection is achieved, but estimation accuracy deteriorates due to zigzag distortion in linear contours
Solution Approach 1:
The patent introduces geometric parameters (curvature, orientation, shape) as additional output dimensions from the neural network. By changing the output space from simple pixel-based edge maps to parameterized geometric representations, the system can enforce geometric constraints during training and inference, thereby eliminating zigzag distortions while maintaining detection accuracy.
Solution Approach 2:
The patent employs a feedback mechanism where estimated geometric parameters are fed back into the system to guide the edge detection process. The neural network iteratively refines its edge predictions by comparing them against the estimated geometric parameters, allowing correction of distortion errors through continuous feedback loops that enforce geometric consistency.
2Shape
If pixel-level classification is performed, then detailed contour information is obtained, but geometric shape accuracy deteriorates
Solution Approach 1:
The patent segments the edge detection task into multiple components: pixel-level classification for contour detail and separate geometric parameter estimation for shape accuracy. By dividing the problem into these distinct segments, the system can optimize each component independently, using pixel-level data for detailed contour mapping while using geometric parameter models to ensure overall shape precision.
Data Source
AI summary
There are included: a learning data acquiring unit to acquire learning data that is a combination of a learning image, a correct edge image indicating a correct edge in the learning image, and a correct geometric parameter related to a shape of the correct edge; an edge estimating unit including a neural network to output an estimated edge image indicating an estimated edge of the learning image and an estimated geometric parameter related to a shape of the estimated edge by inputting the learning image to the neural network; a cost calculating unit to calculate a cost for evaluating estimation accuracy by the edge estimating unit by using the correct edge image, the correct geometric parameter, the estimated edge image, and the estimated geometric parameter; and a model parameter updating unit to update a model parameter in the neural network by using the cost calculated by the cost calculating unit.