Bellaby, GJ, 2000. The use of word level cues for script recognition. PhD, Nottingham Trent University.
Preview |
Text
10290314.pdf - Published version Download (56MB) | Preview |
Abstract
Script recognition systems require the use of context to disambiguate input efficiently. One important consequence of the variability of cursive handwriting is ambiguity. Cues at the word level are shown to influence the performance of human readers. A way to build the factors which produce this superior context effect into a machine system is described. This method of using word level cues is called the word level method.
Different, but complementary, sources of information have been integrated to create a robust and accurate machine system. Specifically, a conventional character based pattern recognizer and a word level method have been successfully amalgamated. Work on the integration of information taken from the meta-word level (semantic and syntactic) is also described.
Integration has provided the opportunity to develop interactive processes within the machine system. This system successfully integrates top-down, and bottom- up, processes. The results of integration for one test sample show an increase in the proportion of target words top ranked from 61% to 70%, in the top 10 from 71% to 83%, and in the top 100 from 72% to 90%. These results demonstrate the efficacy of the word level method and result in a system which has greater scope for improvement when using higher level context. The work described in this thesis is the author's own, unless otherwise stated, and it is, as far- as he is aware, original.
Item Type: | Thesis |
---|---|
Creators: | Bellaby, G.J. |
Date: | 2000 |
ISBN: | 9781369325638 |
Identifiers: | Number Type PQ10290314 Other |
Divisions: | Schools > School of Science and Technology |
Record created by: | Laura Ward |
Date Added: | 06 Jul 2021 10:10 |
Last Modified: | 10 Apr 2024 15:26 |
URI: | https://irep.ntu.ac.uk/id/eprint/43341 |
Actions (login required)
Edit View |
Statistics
Views
Views per month over past year
Downloads
Downloads per month over past year