A printer reverse characterization model based on BP neural network

For colour printer, there are very complicated nonlinear relation between its printed colour chromatic values and input digital image pixel values. In the research, data sets of printed colour chromatic values and their digital image pixel values are classified by hue angle range, the data in each hue angle range is taken as learning samples to create BP neural network. With improved combined method of additional momentum factor and variable learning rate, BP neural network of each hue angle range is trained and created. The experiment result shows that, with appropriate structure and classified learning samples, the reverse characterization model based on ten BP neural networks can be trained in relative short time; the colour errors between the experimental printed colour chromatic values and computed printed chromatic values are far less than the threshold of human eyes, i.e. the reverse characterization model achieves rather high accuracy.