Adaptive clustering model for improving color consistency
Color consistency is one of printing quality issues in color laser printer, a.k.a. color electrophotography (EP) printing system, industry. Maintaining consistent color throughout the cartridge or printer life is a very important engineering challenge for color laser printer manufacturers. In this research, we aimed at developing a new sensor mapping model applied in the printer calibration to improve the color consistency. ^ In order to investigate the possible factors affecting sensor mapping using in the calibration, in this research, time-series sensor data and color measurements have been collected from off-the-shelf color EP printers under a variety of operating conditions. The data analysis shows that the sensor mapping has distinctive behaviors under different levels of relative humidity and cartridge toner consumption. In addition, the sensor mapping has been found to be sensitive to tone level. A new prediction model is proposed to compensate for environmental and consumable disturbances and to capture tone-level-dependent characteristics. The experimental results show the proposed model is able to improve the prediction accuracy by 30% on average. ^ In addition to the environmental and consumable factors, the categorical feature such as cartridge has been observed to be the source of sensor mapping variation. In order to investigate this cartridge impact, a new clustering method with complete must-link (CML) constraints was applied. The supplementary information such as the principal component score is suggested adding into the similarity matrix of clustering algorithm to handle the cartridge clustering problem, one of CML constraint problem. The experimental results not only confirmed the validation of applying supplementary information in the clustering method, but also showed the cartridges can be successfully clustered by their sensor mapping performance. ^ By following the cartridge clustering study, a new sensor mapping model incorporating with cartridge classification module was proposed to improve the sensor mapping. In order to handle the limitation by the nature of calibration, the two-step framework first handled clustering problem by considering variation reduction of sensor mapping, and then developed the classifier by selected features. The data analysis and feature selection were performed to search useful features which are able to classify cartridge group. The experimental results show that classifying cartridges by selected printer sensor information is feasible. Last but not least, the accuracy of sensor mapping can be further improved by ∼ 10% on average if the cartridge classification module is considered.^
Yuehwern Yih, Purdue University.