Inkjet Print Inspection Using Random Dot Pattern Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Inkjet printers face challenges in improving inspection accuracy, particularly when printing various characters and symbols, including Japanese language characters, due to the difficulty in individually learning these characters and the need for efficient inspection operations.
Innovation Solution
A printing inspection device with a machine learning function unit that includes a random test printing function unit and an evaluation function training unit, utilizing a neural network to optimize the evaluation function for inspecting printed images based on a dot arrangement pattern including random dots.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional character-by-character learning method is used for printing inspection, then inspection accuracy for specific characters can be improved, but the system cannot handle the diversity of characters and symbols including Japanese language characters
Solution Approach 1:
The patent uses random dot patterns as simplified copies of character structures for training the neural network. Instead of learning complex actual character forms, the system learns from randomized dot arrangements that capture essential spatial relationships, enabling generalization across diverse character types including Japanese characters.
Solution Approach 2:
The patent transforms the inspection problem by changing the parameter representation from specific character forms to random dot patterns. This parameter transformation allows the neural network to learn invariant features that apply across different character types, resolving the contradiction between accuracy for specific characters and versatility across all character types.
2Measurement precision
If machine learning with fixed character forms is implemented, then inspection accuracy for those specific characters improves, but the system cannot adapt to different characters printed on different printing targets
Solution Approach 1:
The patent makes the training data dynamic by using random dot patterns instead of fixed character forms. This dynamic approach allows the neural network to learn adaptable features that work across different characters and printing targets, eliminating the need for retraining when character types change and improving operational efficiency.
Solution Approach 2:
The random dot pattern training method creates a universal inspection system that can handle multiple character types including Japanese characters, Latin characters, and symbols. The neural network learns general spatial patterns that apply across different character forms, making the system multi-functional without requiring separate training for each character type.
3Measurement precision
If comprehensive character learning is attempted for all possible characters and symbols, then inspection accuracy across all types improves, but the system complexity and training requirements increase significantly
Solution Approach 1:
The patent extracts the essential features of characters by representing them as random dot patterns, separating the critical spatial relationship information from the specific character form details. This extraction simplifies the training data while retaining the essential information needed for accurate inspection, reducing system complexity.
Solution Approach 2:
Instead of providing complete character forms for training, the patent uses partial representations in the form of random dot patterns. This partial action approach provides sufficient information for the neural network to learn inspection capabilities without the complexity of complete character data, achieving good accuracy with reduced system complexity.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
Provided is a novel inkjet printer system that can use machine learning to improve the inspection performance of a printing inspection device. The inkjet printer system includes, in the machine learning, a random test printing function unit (301) that has a function that prints a random dot arrangement that is based on various characters printed in the printing area of a printing target, and an evaluation function training function unit (302) that has a function that optimizes an evaluation function for inspecting a printed image on the basis of a captured image of the results of the printing.