Abstract
An accurate color printer model for nonlinear printer is required to apply flexible gray component replacement (GCR). Conventional method to create color printer model, such as Neugebauer-Yule, high order polynomial, or ordinary neural network do not have enough capability to meet the quality which flexible GCR needs, and as a result, color transformation quality was poor. So, we modified ordinary neural network and successed to make more accurate color printer model. In this paper, the detail of this modification is described.
The effects on predictability by the size of neural network and the number of training patterns (number of printed colors) are also discussed. Our experiment shows the fact that larger network has worse predictability than smaller network when small size of patterns is used.
