Finding Colour Dots by Neural Network

Abstract

We propose a neural network based technique for finding areas of paper covered by ink pigment in coloured half tone newspaper prints. Only one neurone with non-linear output function is used to solve the task. The areas are found by classifying pixels of the image of the print into two classes of colours, namely, the class being considered and the background. The decision boundary between the two classes is found by training the neurone in an unsupervised manner. During the training process the network tries to locate the decision boundary between the two classes in a place sparse in data points. Experimentally we have shown the method to be robust and accurate. We found a good correspondence between the results provided by the method and the results obtained from the particle induced X-ray analysis.