Hierarchical neural network for color classification

One application area of automatic computer analysis of colored images is quality control of multicolored pictures in newspaper printing. The multicolored pictures in newspapers are made by printing cyan, magenta, yellow, and black dots on each other in screens with different angles. During the printing process, the operator needs to control the amount of ink of the different colors to achieve the desired result. One important factor which influences the result is the percentage of the area covered by ink of the different colors in every part of the printed picture. This can easily be determined if one is able to recognize the color of every pixel in computerized image of the print. The authors look at how neural networks of different type and different unsupervised learning techniques were combined to produce a hierarchical architecture with classification accuracy high enough to use in print quality control.<<ETX>>