Zhongliang Fu 
  
  
  
al Lia 
e-— wrens 
  
  
  
  
  
Fig.4e Vertical project Fig.4f Character segmentation 
Experiment results shows that all characters in 24 images is accurately identified, respective only one character 
in two images is falsely identified because these images are partly sheltered, and extraction of characters in 
other two images is difficult because reflex is too strong. Right identification rate reaches 92.8%. If the number 
of images increases, the rate will be improved. By combining automatic identification with manual back-check 
and modification, identification rate may reach 100% to satisfy the demand in applications. For every image, the 
time spent is less than 0.2 second. 
REFERENCES 
1. Andreas Uert and Jung Yang, Extraction of Text and Symbol from Unstructured Vector Data, Hannover, 
1987,3 
2. Keiji Yamada, Handwritten Numeral Recognition by Multi-layered Neural Network with Improved Learning 
Algorithm. International Joint Conference on Neural Networks, Washington, D.C. Jun,18-22,1980 
3. M. Karnel and A. Zhao, Extraction of Binary Character/Graphics Images from Grayscale Document Images. 
Graphic Models and Image Processing, 55, 1993, 203-217 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 311