Gregory Fraser Russell

Handwritten document retrieval

This paper investigates the use of both typed and handwritten queries to retrieve handwritten documents. The recognition-based approach reported here is novel in that it expands documents in a fashion analogous to query expansion: Individual documents are expanded using N-best lists which embody additional statistical information from a hidden Markov model (HMM) based handwriting recognizer used to transcribe each of the handwritten documents. This additional information enables the retrieval methods to be robust to machine transcription errors, retrieving documents which otherwise would be unretrievable. Cross-writer experiments on a database of 10985 words in 108 documents from 108 writers, and within-writer experiments in a probabilistic framework, on a database of 537724 words in 3342 documents from 43 writers, indicate that significant improvements in retrieval performance can be achieved. The second database is the largest database of on-line handwritten documents known to its.

Ink Retrieval from Handwritten Documents

This paper compares several information retrieval (IR) methods applied to the problem of retrieving specific words from a handwritten document. The methods compared include variants of the Okapi formula and Latent Semantic Indexing (LSI); recognition-based retrieval; and keyword search. One novel aspect of the work presented is that it uses the output stack of a Hidden Markov Model (HMM) handwriting recognizer with a 30,000-word lexicon to convert each handwritten word into a document which is then used for document retrieval. Preliminary experiments on a database of 1158 words from 75 writers indicate that the keyword search has superior precision and recall for text queries, and that ink queries result in minor performance reductions.

Retail applications of signature verification

The dramatic rise in identity theft, the ever pressing need to provide convenience in checkout services to attract and retain loyal customers, and the growing use of multi-function signature captures devices in the retail sector provides favorable conditions for the deployment of dynamic signature verification (DSV) in retail settings. We report on the development of a DSV system to meet the needs of the retail sector. We currently have a database of approximately 10,000 signatures collected from 600 subjects and forgers. Previous work at IBM on DSV has been merged and extended to achieve robust performance on pen position data available from commercial point of sale hardware, achieving equal error rates on skilled forgeries and authentic signatures of 1.5% to 4%.