Similarity-based training set acquisition for continuous handwriting recognition
Information Sciences: an International Journal
| Hi-index | 0.00 |
This paper describes three types of noise that occur during the scanning process and presents pre-processing techniques to remove them from handwriting images while retaining the details of the handwriting. In our application (forensic document examination) the details of handwriting need to be retained to ensure its admissibility as an evidence in a court of law, while in character recognition applications the details of the handwriting are not considered important as long as their absence does not affect the ability of the classifier in correctly recognizing those letters. Thus some pre-processing methods that are acceptable in character recognition are not acceptable in forensic document examination. This paper also analyses the behavior of shadow noise, which has not received in character recognition. This paper also shows that different types of inputs require different pre-processing methods.