Signature and lexicon pruning techniques

Handwritten word recognition and signature identification are important areas in machine vision that require extensive exploration for improved results. The performance of such systems tend to degrade when the number of choices to be dealt with increase. In case of a lexicon-driven handwritten word recognizer, the performance degrades when the lexicon size increases [H. Xue and V. Govindaraju, 2002]. Similarly, the matching process is tedious when the number of reference templates increase in case of online signature identification. For better performance, both in terms of recognition rates and response time, interactive models are suggested, which involve feedback process to further enhance the systems. Interactivity is attained by choice pruning, which filter out useless entries, thus providing the system with a smaller set for further detailed investigation. The paper mainly identifies the necessity for choice pruning and deals with two specific cases - signature pruning and lexicon pruning.