By applying the above processes, two characters that have been detected as the same orientation are selected, and the orientation of the scanned document image is determined to process the same orientation as these two character images. As a result of this, the accuracy of orientation detection is improved 100%.
Applying these algorithms to sample documents chosen from magazines, newspaper and other ordinary paper documents, correction of document orientation achieves 100% accuracy. This technique can be applied to any document image, except for picture or graphical images which don't include any characters in the image. The correction processing takes about 0.4 seconds on an ordinary personal computer with an MMX Pentium-233 MHz CPU. It is short enough for practical use.
This technique dramatically reduces user operations on the computer before and after scanning, and makes scanning operation easier, so that user can manipulate the scanner in any direction and always obtain the correct document image.
Kamada, H, Fujimoto, K. & Akimoto, H.( 1997). "High-speed, High-accuracy Document Recognition Method using Compressed Character Features and Original Character Features". Technical Repot of IEICE, PRMU96-205, 97-104.
Fujimoto, K. & Kamada, H.( 1996). "Fast and Precise Character Recognition by Estimating Recognition Probability". Proceedings of the 1996 Engineering Science Society Conference of IEICE, D-361, 364.