Collaborative Cross-media Annotation of Documents - Human-Computer Interaction - page 122

Fig. 5.14 Example of an annotation with text which is difficult to recognize
Fig. 5.15 An annotation that combines text with graphics
Improving the Accuracy of Handwriting Recognition
Given these problematic aspects, we aimed at improving the recognition accuracy
of domain-specific terms. This is important because domain-specific terms are often
used in search queries as well as for tagging and indexing documents.
For these experiments, we removed annotations that contain drawings from our
test set. This results in a total of 118 purely textual annotations. A considerable
percentage of 11.3 % of all terms in these annotations are domain-specific terms.
The selection of domain-specific words 3 from the field of computer science and the
3 The following terms, contained in the slides and/or the annotations, were classified as
domain-specific (slides were in German): Pivot, Pivotelement, Worst Case, Sortiert, Quicksort,
O(nlogn), O(n), O(n2), Zeiger, Daten, Baumstruktur, AVL-Baum, Schl ussel, Sortierung, Zwis-
chenordnung, Durchlaufe, Durchlauf, Speicherbedarf, Speicher, Binarbaum, Array, Suchbaum,
Wurzel, Knoten, Sortieren, Sortierschritte, Mergesort, Heap, Heapbedingung, Heapsort, Laufzeit,
Aufwand,Einsinken, tauschen, austauschen, Liste, Teilliste, optimieren, Sortieralgorithmus, Ele-
ment, Java, Bottom-Up-Verfahren, Datenbl ocke, Treesort, Out-of-Place, Sortierverfahren, Permu-
tation, Datenmengen, Cluster, Inordertraversierung, Pseudocode, Datenstruktur, Rekursion, binar,
Logarithmus, Heapbereich, rekursiv.