Contour Line Prediction for Faster Training Image Annotation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for creating training images for machine learning require significant manual labor and time, especially when dealing with images containing complex contour lines, and existing annotation tools are inadequate for handling such images.
Innovation Solution
An image processing apparatus that includes an image acquiring unit, a contour line predicting unit, and an image output unit, utilizing a learned model to predict contour lines in target images, thereby automating the creation of training images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If manual operations are used to create training images, then annotation tools can handle simple images, but the process requires large amounts of labor and time and cannot handle images with complicated contour line shapes
Solution Approach 1:
The patent replaces manual mechanical annotation operations with an automated image processing system that uses machine learning models to predict contour lines. The system substitutes human labor with computational algorithms that can automatically identify and annotate complex contour lines in images, thereby handling complicated shapes without increasing labor requirements.
Solution Approach 2:
The system enables self-service annotation by training a machine learning model to automatically predict contour lines without human intervention. Once trained, the model can independently process images and generate annotations for complex contour lines, making the system self-sufficient for handling various image types including those with complicated shapes.
2Productivity
If automated annotation tools are used, then labor and time are reduced, but these tools cannot handle images with complicated contour line shapes
Solution Approach 1:
The patent applies preliminary action by training a machine learning model in advance with diverse training data that includes images with complicated contour line shapes. This pre-training enables the automated annotation tool to handle complex shapes when deployed, resolving the limitation of conventional automated tools that cannot process complicated contours.
Solution Approach 2:
The system changes the parameters of the annotation approach by using machine learning models with adjustable parameters that can be optimized for different image types. By tuning the model parameters and using appropriate loss functions during training, the system achieves the capability to detect and annotate complex contour line shapes while maintaining automated efficiency.
3Measurement precision
If more training data is collected to improve machine learning model accuracy, then image analysis accuracy improves, but the amount of manual work required increases
Solution Approach 1:
The patent uses copying by generating synthetic training data through data augmentation techniques. Instead of manually collecting diverse real images, the system creates copies and variations of existing training images through transformations such as rotation, scaling, and flipping, thereby expanding the training dataset without proportional increases in manual work.
Solution Approach 2:
The system applies segmentation by dividing the training data creation process into manageable components. It processes images in segments and uses automated annotation on portions of images, combining these segmented results to create comprehensive training datasets, thereby reducing the overall time required compared to manual annotation of complete images.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
An image processing apparatus includes an image acquiring unit that acquires a target image indicating a target position specified by a user; a contour line predicting unit that predicts a contour line near the target position based on a learned model that has learned a relationship between a position in an image and a contour line near the position; and an image output unit that outputs a training image indicating a prediction result of the contour line near the target position.