Neural Network Color Prediction Model for Digital Printing

Solution Overview

Problem

The existing methods for predicting colors obtained by overprinting inks of multiple colors, such as the Deshpande et al. method, require printing and colorimetry of multiple patches in a CxF chart, leading to increased costs and man-hours, especially when using spot colors in digital printing.

Innovation Solution

A color prediction model is constructed using machine learning by a neural network that learns the relationship between spectral characteristics of reference patches and color values, allowing for the prediction of color overprints without the need for actual printing and colorimetry, thereby reducing costs and man-hours.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If the Deshpande et al. method is used to predict colors obtained by overprinting inks, then overprint prediction values can be calculated using linear equations, but the cost and man-hours increase due to the need to print and perform colorimetry on multiple patches in a CxF chart

Engineering Contradiction:

Improvecolor prediction accuracyVSAvoidman-hours for printing and colorimetry

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent applies preliminary action by pre-calculating and storing overprint prediction values for all possible combinations of spot color inks and their respective dot percentages in a lookup table during system initialization. When color prediction is needed, the system simply retrieves pre-computed values based on the input parameters, eliminating the need for real-time printing and colorimetry measurements.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent creates a virtual color chart through computational modeling that replicates the functionality of a physical CxF chart. Instead of physically printing patches and measuring them, the system uses a lookup table that contains copied measurement data from previous calibrations, allowing color prediction without physical prototypes.

Inventive Principle:

Principle #26Copying

2Measurement precision

If the Deshpande et al. method is used to predict colors obtained by overprinting inks, then overprint prediction values can be calculated using linear equations, but the cost increases due to the need to print and perform colorimetry on multiple patches in a CxF chart

Engineering Contradiction:

Improvecolor prediction accuracyVSAvoidcost of printing and colorimetry materials

Core Design Contradiction:

Measurement precisionVSLoss of substance

Solution Approach 1:

The patent applies preliminary action by pre-calculating and storing overprint prediction values for all possible combinations of spot color inks and their respective dot percentages in a lookup table during system initialization. When color prediction is needed, the system simply retrieves pre-computed values based on the input parameters, eliminating the need for real-time printing and colorimetry measurements.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent creates a virtual color chart through computational modeling that replicates the functionality of a physical CxF chart. Instead of physically printing patches and measuring them, the system uses a lookup table that contains copied measurement data from previous calibrations, allowing color prediction without physical prototypes.

Inventive Principle:

Principle #26Copying

3Measurement precision

If a CxF chart is printed and colorimetry is performed to obtain accurate color values, then color prediction accuracy is improved, but the process requires significant man-hours and cost

Engineering Contradiction:

Improvecolor value accuracyVSAvoidprediction process efficiency

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The patent applies preliminary action by pre-calculating and storing overprint prediction values for all possible combinations of spot color inks and their respective dot percentages in a lookup table during system initialization. When color prediction is needed, the system simply retrieves pre-computed values based on the input parameters, eliminating the need for real-time printing and colorimetry measurements.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent creates a virtual color chart through computational modeling that replicates the functionality of a physical CxF chart. Instead of physically printing patches and measuring them, the system uses a lookup table that contains copied measurement data from previous calibrations, allowing color prediction without physical prototypes.

Inventive Principle:

Principle #26Copying

Data Source

PatentEP3675476B1Color prediction model construction method and color prediction model construction program

Publication Date: 2023.11.15 SCREEN HOLDINGS CO LTD

Figure 1~2
Figure 3
Figure 4

AI summary

A neural network which takes the spectral reflectances of a reference patch as input data and takes the spectral reflectances of a color prediction target patch as output data is prepared. Teaching data for learning is acquired (S100), and the relationship between the spectral reflectances of the reference patch and the spectral reflectances of the color prediction target patch is learned by the neural network (S110) using the teaching data. The spectral reflectances of the reference patch for a color prediction target ink are measured (S120), and the spectral reflectances which are the measurement results are given to the learned neural network (S130). Thereby, the spectral reflectances of the color prediction target patch for the color prediction target ink are predicted.