Color Prediction Model Transfer Learning for Printing Mediums
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing techniques for creating color prediction models in printing devices are inefficient and inaccurate, especially when dealing with a large number of ink colors, as they require extensive time and effort to print and measure color charts on various mediums, and are affected by the texture and tension of fabrics, leading to suboptimal color measurement data.
Innovation Solution
A color prediction model creation device and system that utilize a first color prediction model learned from a first printing medium to create a second color prediction model for a different printing medium with fewer ink combinations, reducing the need for extensive color measurement and printing efforts by employing transfer learning and a sublimation transfer process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the number of ink colors used in a printing device increases, then the color reproduction capability is improved, but the number of combinations of ink amount sets increases enormously, leading to increased time and effort required for printing a color chart and measuring colors
Solution Approach 1:
The system performs preliminary color measurement and data collection for a first printing medium, then uses this pre-acquired data as a foundation for creating color prediction models for second printing mediums. This preliminary action reduces the need to collect extensive data for each new medium, thereby reducing time and effort.
Solution Approach 2:
The system creates a color prediction model for a second printing medium by copying and adapting the model structure and learned parameters from a first printing medium. Instead of building a completely new model from scratch for each medium, the system copies the foundational model and fine-tunes it with limited new data, significantly reducing measurement requirements.
2Adaptability or versatility
If color measurement is performed on a fabric product with texture and tension variations, then the color prediction model can be created for the final product, but the measurement accuracy deteriorates due to texture and tension affecting the color data
Solution Approach 1:
The system introduces an intermediate transfer paper as a mediator between the printing device and the fabric product. Color charts are printed on this smooth, stable transfer paper where measurements can be accurately taken, avoiding the texture and tension issues of fabric. The color prediction model then learns the relationship between ink amounts and colors through the sublimation transfer process to the fabric.
Solution Approach 2:
The measurement process is segmented into two distinct stages: first, accurate color measurement on a stable transfer paper medium; second, application of the learned model to predict colors on the fabric product. This segmentation allows high-precision measurements to be performed on a suitable medium while still achieving the goal of fabric color prediction.
3Measurement precision
If a color prediction model is created by learning from a large number of ink amount combinations, then the model accuracy is improved, but the number of color charts that must be printed and measured increases, increasing time and effort
Solution Approach 1:
The system performs preliminary learning on a first printing medium to establish a foundational color prediction model. This preliminary learning phase captures the general relationship between ink amounts and colors, which can then be transferred to second printing mediums with fewer additional measurements required, improving efficiency while maintaining accuracy.
Solution Approach 2:
The color prediction model structure, learned parameters, and relationships are copied from the first printing medium to the second printing medium. This copying approach allows the system to achieve good model accuracy for the second medium with significantly fewer measurement data points, as the foundational knowledge has already been acquired through preliminary learning.
Data Source
AI summary
A color prediction model creation device configured to create a color prediction model for predicting color values of a device-independent color system from an ink amount set includes: a first color prediction model creation unit configured to acquire a first spectral reflectance of a color chart printed on a first printing medium by using a first ink amount set, and create a first color prediction model for the first printing medium by learning a correspondence relationship between the first ink amount set and the first spectral reflectance; and a second color prediction model creation unit configured to acquire a second spectral reflectance of a color chart printed on a second printing medium by using a second ink amount set having a smaller number of combinations than that of the first ink amount set, and create a second color prediction model for the second printing medium by learning using the second ink amount set, the second spectral reflectance and the first color prediction model.