See for another approach:
http://www.ai.rug.nl/alice/nwo-catch-scratch/index_english.html. From that page: For one thing, the massive amount of text images of handwritten pages will allow for the exploitation of modern statistical techniques. Data mining techniques such as clustering will uncover regularities in handwritten shapes that may act as the bridge to document retrieval and text analysis. In this project we will not aim for a veridical left-to-right transcription of handwritten documents: this would be an unrealistic target. Alternatively, at the end of the project, we will have delivered tools for keyword-based search in handwritten archives, akin to existing flat-text search methods ("Googling").