Adapting BLSTM Neural Network Based Keyword Spotting Trained on Modern Data to Historical Documents

Abstract

Abstract-Being able to search for words or phrases in historic handwritten documents is of paramount importance when preserving cultural heritage. Storing scanned pages of written text can save the information from degradation, but it does not make the textual information readily available. Automatic keyword spotting systems for handwritten historic documents can fill this gap. However, most such systems have trouble with the great variety of writing styles. It is not uncommon for handwriting processing systems to be built for just a single book. In this paper we show that neural network based keyword spotting systems are flexible enough to be used successfully on historic data, even when they are trained on a modern handwriting database. We demonstrate that with little transcribed historic text, added to the training set, the performance can further be enhanced.

A Novel Word Spotting Method Based on Recurrent Neural Networks

“…8 the effect of the adaptation to the GW data set can be seen as a reduction of the label error rate. This approach with similar experiments has also been proposed in [44].…”

“…A preliminary version of the system described in this paper has been presented in [43], [44]. The current paper provides significant extensions with respect to the underlying methodology and the experimental evaluation.…”

“…As the second extension over [43], [44], an extensive comparison with several reference systems is presented. On the one hand, a common DTW algorithm as well as a modern HMM-based algorithm is used for comparison.…”

BLSTM Neural Network Based Word Retrieval for Hindi Documents

“…This category of neural networks was previously used for speech recognition [5], handwriting recognition [6], as well as keyword spotting for English handwritten text [7]. We extend these methods for wordspotting in printed Hindi documents.…”

“…Another common technique for sequence recognition are Hidden Markov Models (HMMs). However, on a similar keyword spotting task they are outperformed by BLSTM based keyword spotting systems [7].…”