Automated Ground Truth Mask Generation for Image Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The generation of pixel-perfect ground-truth training data for image segmentation applications is difficult and costly due to the need for extensive human annotation, especially for machine learning approaches that require annotated data.
Innovation Solution
A computer-implemented method that generates training data by parsing input documents to identify predefined items, creating bitmap images, and assigning encoded class labels to pixels based on position information, allowing for efficient generation of masks that can be used to enhance training data sets for image segmentation applications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If human annotation is used to generate pixel-perfect ground truth, then the quality and accuracy of training data is improved, but the cost and time required increases significantly
Solution Approach 1:
The patent uses parsed document structure data (bounding boxes, text content, layout information) to automatically generate pixel-perfect ground truth masks by copying and transforming the structured data into pixel-level annotations, eliminating the need for manual pixel-by-pixel annotation while maintaining high accuracy
Solution Approach 2:
The system performs self-annotation by automatically parsing documents, identifying text and layout elements, and generating ground truth masks without human intervention, allowing the data generation process to serve itself rather than requiring external annotators
2Measurement precision
If human annotation is used to generate pixel-perfect ground truth, then the quality and accuracy of training data is improved, but the cost increases significantly
Solution Approach 1:
The patent transforms structured document data (which is already available from parsing) into pixel-level ground truth masks through automated copying and transformation processes, eliminating costly manual annotation while preserving the precision needed for training machine learning models
Solution Approach 2:
The system automatically generates its own training data by parsing documents and creating ground truth masks without human annotators, dramatically reducing the cost of data production while maintaining high accuracy through algorithmic precision
3Adaptability or versatility
If machine learning approach is used for document component extraction, then the generality and adaptability is improved, but the requirement for annotated training data increases
Solution Approach 1:
The patent performs preliminary parsing of documents to extract structured information (text, bounding boxes, layout elements) before generating training data, creating a foundation that enables automatic ground truth generation and reducing the need for extensive manual annotation of training sets
Solution Approach 2:
The system copies parsed document structure data to create pixel-level ground truth masks, generating the training data needed for machine learning models through automated transformation rather than manual annotation, thereby providing the quantity of training data required without proportional increase in cost
Data Source
AI summary
A method, system and computer program product to generate a training data set for image segmentation applications, comprising providing a set of input documents of a first format. The input documents each comprise one or more pages. The input documents are split into individual document pages and parsed. Parsing comprises identifying a predefined set of items including position information of the position of the predefined set of items in the individual document pages; generating a bitmap image of a second format for each individual document page of the first format. The bitmap image comprises a predefined number of pixels. A mask is generated for each individual document. The mask comprises the predefined number of pixels of the corresponding bitmap image. Generating the mask comprises assigning an encoded class label to each pixel of the mask based on the position information of identified items of the predefined set of items.