A Novel Approach to Extract Digital Signatures from Handwritten Character Images

Manuscript character recognition is undoubtedly an interesting and practical problem. While the current state of the art solutions are very effective, they mostly rely on Artificial Neural Network based models for performing characters recognition. Such models typically require significant amount of training data in order to produce high accuracy results. We are proposing a method for extracting the most significant features of character images and creating a low dimensional space numeric signature for each character image shape. Such signature can be used to measure similarities between character images, and subsequently perform clustering, classification or searching for character images. Our methodology is suitable to manuscript character images, for which, typically, there are no pre-trained advanced models that can be used to perform clustering or classification tasks. It can be used as a pre-classification step to perform automatic labeling of training data.