Technical image reduction using NN and wavelet

A general-purpose procedure for scaling technical line drawings, suitable for video presentation, is described. The proposed method is based on the separate processing of scalable (layout) and non-scalable (symbol) elements, drawn from standard technical drafting symbols, detected by a cluster-based template procedure and a minimum distance classifier, are extracted from drawings and utilized to form a symbols position table. To obtain the clusters of symbols, a rival penalized competitive learning neural network and a human template labeling procedure have been adopted. The extraction of symbols from drawings produces clear layouts. These layouts are scaled down by wavelet based algorithm and the symbols are then restored or replaced, through the symbols position table, with different graphs or textual representations, according to the scaling factors and the display device. The results of an experimental study on a large database of technical drawing are presented and the accuracy of the system is discussed