Automated document image preprocessing management utilizing grey-scale image analysis and neural network classification

This paper describes a new approach for preprocessing grey-scale document images. The grey-scale image is first partitioned and analyzed using a set of image features to extract the quality and contents information in the given entity. Then, this data is classified using a neural network classifier to find and prioritize the need for image enhancement, if any. The classification information guides a special filter bank and their parameters to perform prefiltering operation to document image. In filtering, a control method is used to assure that no overfiltering occurs. The system behaviour and experiments are presented showing the performance of the developed techniques with tests performed with three different document databases