Synthesized Defect Images for Digital Printing Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current inspection methods for printed materials struggle to accurately determine the origin of defects in printed designs, as they rely on comparing acquired images with golden masters, which are often of insufficient quality and require manual judgment, and cannot efficiently isolate defects caused by specific nozzle malfunctions in digital printing presses.
Innovation Solution
A method and system using trained synthesis neural networks to generate synthesized defective images that mimic potential defects, allowing for the identification of defect origins by comparing these images with acquired images, thereby reducing the number of images needed for analysis and improving defect localization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If golden master images are used for defect detection, then defect detection can be performed, but the quality is insufficient and manual judgment is required
Solution Approach 1:
The patent creates virtual copies of defective images by synthesizing defects into reference images. Instead of relying on manual creation of golden masters, the system automatically generates multiple versions of reference images with different defect types and locations, enabling automated defect detection without manual judgment while maintaining high accuracy.
Solution Approach 2:
The patent performs preliminary synthesis of defective reference images before actual defect detection. By pre-generating images with various potential defects at different locations and types, the system prepares all possible defect scenarios in advance, eliminating the need for manual judgment during actual inspection and improving both accuracy and automation.
2Measurement precision
If multiple images are used for defect analysis, then defect origin can be identified, but the number of images required is large
Solution Approach 1:
The patent segments the defect identification process by creating separate synthesized images for different defect types and locations. Each synthesized image represents a specific hypothesis about the defect origin, allowing the system to identify the correct origin by comparing with the actual defect image without needing all possible images simultaneously, thus reducing the effective number of images needed for analysis.
Solution Approach 2:
The patent generates multiple synthesized images with defects at different locations and types, but only a subset of these images is needed to identify the defect origin. By creating more images than strictly necessary and then selecting the best matches through comparison, the system ensures accurate identification while the actual processing only requires analyzing the relevant subset of synthesized images.
Data Source
Figure 1
Figure 2A~2D
Figure 2E
AI summary
Method for generating a test-set for inspection of a design being printed by a printing-press, each color-unit in the printing-press prints a respective color. The design is composed of original-layers. Inspection includes determining the origin of at least one defect in the printed-design. The method includes the procedures of generating defective-layer or layers of the design, by introducing at least one selected defect to at least one selected original-layer, in a selected location or locations and combining layers using a trained-synthesis-neural-network. The layers include the defective-layer or layers and remaining ones of the original-layers. The trained-synthesis-neural-network provides a plurality of features respective of each pixel. The method also includes the procedure of generating the test-set from the output of the synthesis-neural-network. The test-set includes at least one synthesized-test-image. The synthesized-test-image includes at least one synthesized-defect at the selected location. The test-set is employed to determine the origin of the defect.