The Research And Application Of Fax Recipient's Name Recognition Technology

Fax recipient's name auto-recognition system belongs to the area of special OCR system development. The development of special OCR system will widen the application field of OCR technology in great degree and improve the automation degree of relative application field. Meanwhile, some new problems and new methods encountered in the system implementation process have some theoretical value for the research of OCR technology as the object processed by the system is special. In special OCR system, the technology of fax recipient's name recognition has considerable application future and practical value.The fax recipient's name recognition technology is researched and applied in this paper. In the aspect of characters'image pretreatment, according to the speciality of fax recipient's name characters image, some methods such as the binarization, denoising, skew correction, size normalization and thinning appropriate for the fax characters image are adopted besides the research of traditional method in this paper; and the analyze is implemented for the detection of underline and removing method as it is possible to have underline under fax recipient's name characters, and a method based on mathematical morphology for removing underline is utilized; and utilizing crest-trough analyses, the method for dividing characters divides characters'image into single characters, and at the same time, a dividing method according to the information of characters strokes to seek for adapted dividing points and paths is applied in this paper because of conglutinated characters.In the aspect of characters recognition, based on their own features of different network models and using multiple neural networks integration technology, a recognition system applying multiple hybrid neural networks with multiple features is built to deal with these problems about characters recognition and the recognition accuracy is improved in this paper. The classifier of recognition system is composed of rough classification and sub-classification. The self-organizing clustering network in this paper can work out the probability distributing estimation of the pattern in multi-dimension space, and then, combining some features such as the index of stroke complexity of characters, four-side codes and rough contour, estimate the result of rough classification of characters. BP network has excellent nonlinear mapping feature and classifying capability for inputted vectors, thus the sub-rank classifier for...