| 摘要 |
<p>The invention relates to a method of manuscript recognition, and to a manuscript recogniser. The method includes, for cursive script recognition, receiving digitised data indicative of manuscript to be recognised, processing the digitised data to derive the manuscript shape in encoded form as a plurality of vector directions, determining from the digitised data a respective quantised length for each vector direction, deriving an encoded vector string indicative of the vector directions and their lengths, comparing sub-string segments of the encoded vector string with stored encoded data representative of ligature shapes and selecting identified sub-string segments indicative of possible ligature shapes, combining each identified sub-string segment with immediately adjacent sub-string segments in the encoded vector string to derive longer sub-string segments and determining the quantised length of a sub-string as a proportional length of a total length of the sub-string defined by the manuscript shape. Each of the longer sub-string segments are compared with stored encoded data representative of character shapes to identify possible matching character shapes, and a letter net is derived indicative of possible character routes through the encoded vector string. The combination of characters in each character route is then compared with known combinations of stored characters to identify the script.</p> |
