Abstract:
An algorithm for indexing and retrieving trademark images using hierarchical stages isproposed in this research to improve searching time over conventional single stageretrieval algorithm, which is significant when dealing with huge trademark database. Anindexing procedure extracts and stores background and foreground shape features oftrademark images which are circularity, rectangularity, eccentricity, hole area ratio andinvariant moments into specific table. The searching procedure is consisted of twostages. First stage, the searching uses geometric features and invariant moment ofbackground component to coarsely select the candidates. If no searching result has beenfound, or too many images are retrieved, then the obtained candidates are used toprocess in the second stage. The second stage searching using foreground shape featuresto compute the result, which is more complicated and has higher accuracy. The resultscan be obtained by using similarity measurement of Euclidean distance on multiplegeometric features. The experimental results with five classes trademarks ranged from2,000 to 10,000 trademarks indicated that the hierarchical stages searching algorithmreduce the average searching time at least five times, comparing to the single stagesearching method. An average recall is 62% and precision is 86%.