Segmentation Model Training With Overlay Composite Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing systems face challenges in efficiently training segmentation models due to the scarcity of sufficient training data, particularly for differentiating between image backgrounds and foregrounds, especially when source image data is not available.
Innovation Solution
A computer-implemented method that augments image data using overlay images to generate composite data for training a segmentation model, allowing the model to differentiate between foreground and background, even in scenarios where source image data is lacking, by applying overlay image data to identified sections of the image.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If real training data with backgrounds and foregrounds is collected, then the segmentation model can be trained accurately, but the availability of sufficient training data is limited
Solution Approach 1:
The patent creates synthetic training data by copying and combining overlay image data with source image data to generate composite images. This copying approach generates additional training samples without requiring more real-world captured data, directly addressing the scarcity of training data while maintaining segmentation training capability
Solution Approach 2:
The patent performs preliminary data preparation by pre-processing overlay images and source images into composite training data before model training. This preliminary action creates a ready-to-use training dataset that eliminates the need for collecting and preparing real training data during the training phase
2Measurement precision
If the segmentation model is trained with sufficient data, then the model can differentiate between foreground and background effectively, but the training time and computational resources increase
Solution Approach 1:
By copying and recombining existing overlay and source images to create multiple composite training samples, the patent generates sufficient training data without requiring proportional increases in data collection time or computational processing time for new data acquisition
Solution Approach 2:
The patent creates a universal training data generation approach where the same overlay and source images can be combined in multiple ways to produce diverse training samples, allowing one set of input images to serve multiple training purposes without additional data collection
3Adaptability or versatility
If overlay image data is applied to generate composite data, then additional training data is generated for scenarios where source image data is not available, but the complexity of the training process increases
Solution Approach 1:
The patent introduces composite image data as an intermediary training format that bridges the gap between available overlay/source images and the segmentation model's training requirements. This intermediary representation simplifies the training process by providing a standardized input format that the model can process directly
Data Source
AI summary
Aspects of the present invention relate to a computer-implemented training method for training a segmentation model to segment an image. The method includes receiving a training data set which includes first image data representing a first image comprising a foreground and a background. The first image data is captured by at least one first visible electromagnetic radiation imaging device. The method comprises augmenting the first image data by applying a first overlay image data to the first image to generate first composite image data. The first composite image data is processed using the segmentation model. The segmentation model is updated in dependence on the processing of the first composite image data to generate a first updated segmentation model. According to a further aspect of the present invention there is provided a system for training a segmentation model to segment an image. Aspects of the present invention also relate to an image processing system and method.