A Printing Dot Area Prediction Model Based On Spectrum

Nowadays,Offset printing is the most common method.Improving the quality and reducing cost with high printing efficiency are the keys of making profit for industry.Ameliorating color reproduction,which is one of evaluations of printing quality,is a factor of great significance for making the better quality.In contrast with the mature color matching system of spot colors,halftone prediction models lack ample researches.In this paper,K-M theory in printing and dyeing industry is quoted to relate spectral reflectance of targets to corresponding tone values of overprinted base colors.This paper propose a model,which is based on BP neural network whose input is the 31-D spectral reflectance vector of targeting color and output is the CMYK values of corresponding color,to predict tone value by spectral reflectance.To assess the model performance,the mean square error(MSE)between training samples and corresponding output of model is applied for stability of model,and the mean square error(MSE1)between testing samples andcorresponding output of model.Finally,calculate color difference as an evaluation for the model after coverting color space.In this paper,firstly complete the collection of experimental data.And then,structure the prediction model.After that,the prediction model can be improved as the following steps,which are optimizing training samples,fine-tuning the hidden layers and layer nodes of neural network,experimenting the initial threshold and weight,ordering the activation and preventing from overfitting.Finally,input testing samples into prediction model,convert the output to CIE L*a*b*and calculate the color difference between output and correspondence.The experimental results show that the average difference is 2.35,which can meets printing quality requirements.Therefore,the halftone prediction model proposed in this paper achieves outstanding stability and accuracy,which can be implemented in production of offset pringting.