期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：1992
卷号：XXIX Part B2
页码：345-349
出版社：Copernicus Publications
摘要：Image binarization is a fundamental research theme in image processing and an improtant preprocessingmethod in image recognition and edge/boundary detection. It is very difficult to select the correspondingthreshold for each image in different application domains. In this paper, we used a multi-layer feed-forwardneural net (multi-layer perceptron) as a threshold transfer to select the visually satisfied threshold by backpropagation algorithm. In order to improve network performance and reduce the occurrence of local minima,we introduced extra hidden units and made many training runs starting with different sets of randomweights. Besides, we also introduced the collective decision produced by the ensemble to less error probablymade by any of the individual network. The architecture of our neural percept ron makes it available toperform multi-stage image processing, programmingly arrange the. relationship between groups within differentlayers, and emphasize not only the local image texture but the globle information. The comparisonwith other threshold-selecting methods shows that our neural net method is able to select visually satisfiedthreshold and to obtain good restoration image using the result binarized image.
关键词：Image Preprocessing; Binarization; Back Propagation; Neural Network
