Use this URL to cite or link to this record in EThOS:
Title: Automatic recognition of ancient Syriac handwriting
Author: Fernando, Pathirajaliyanage Prem Jayalath
Awarding Body: University of Cambridge
Current Institution: University of Cambridge
Date of Award: 2005
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Restricted access.
Access from Institution:
In our experiments we use over 10000 images taken from scribe-written historical manuscripts, from a 19th century typeset document and from other sources. The method we implement, test and describe here does not need to find, strokes or contours of the characters, but exploits a characteristic measure of the shape of the character. The measure is calculated by defining a new Generalized Auto-Regressive Moment Function (GARMF) for each character shape model. Using our GARMF method we present a model to map the input image onto multiple feature planes that are then mapped using a standard predefined kernel function onto a higher dimension plane where linearly separable hyperplanes exist. We implement a comprehensive system, from document skew correction, through connected component labelling and segmentation in word and character levels, to model selection, evaluation and recognition. Novel techniques introduced include methods for document skew correction, word and character segmentation, and the GARMF method. The GARMF method is used to extract feature points specific to the shape of the characters that are tolerant to shape variations, distortions and image quality. Each shape is then recognized individually using a discriminative support vector machine. The model selection is achieved by a 10-fold cross-validation. The resultant model is tested against various other models, moment generating methods and varying combinations of features and SVM models. Confusion matrix and ROC analysis are used in model evaluation. Recognition results approach 100%.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available