We propose the use of neural network for segmenting character,natural image, and screened halftone image documents. The model networkhas learned the characteristics of Laplacian histogram of each document.This network consists of three sub-networks and a decisional circuit.The structure of each sub-network is feed forward four layers structuredneural network consisting of 10×10 input cells, 20 first hiddencells, five second hidden cells and two output cells. Those sub-networkare the one dividing the documents of character and natural images, theone dividing the documents of natural images and screened halftoneimages, and the last one dividing the documents of character andscreened halftone images. Input patterns to sub-networks composed of10×10 data of optical density of pixels having 8 bits' graylevels. The data were processed to have 10% saturation characteristicsto the full scales for compressing scratch noises. Furthermore,Laplacian and logarithm operation were adopted to emphasize differencesof luminosity distribution between character, natural images, andscreened halftone images. An error back propagation method was also usedas a learning rule for the sub-networks. As a result, the total networkpresented successfully more than 90% accuracy of discrimination for eachtypes of document
展开▼