Názov:Comparison of Handwritings
Vedúci:doc. RNDr. Milan Ftáčnik, CSc.
Kµúčové slová:handwriting, writer verification, grapheme, clustering
Abstrakt:The goal of this paper is to contribute to the solution of a following problem. Having one and more scanned images with handwritten text as input, our task is to create methods which can determine whether the documents are written by the same person or not. We concentrate on a fundamental problem - comparison of two images and decision whether two images are written by the same person or not - so-called writer verification problem. We present three approaches consisting of preprocessing, feature vector extraction and combination with graphemes clustering. The first approach is based on feature vector, the second is combination of the first and Kohonen's Self-Organizing Map and finally the third joins the first approach and modified hierarchical clustering. We made our experiments on 100 images from 40 different writers. These images were taken from the IAM Handwriting Database which contains 1539 pages of scanned text from 657 writers. In these experiments we have tested three approaches mentioned on two input images. We achieved 96,5% accuracy by the third approach. A working implementation is presented together with experimental results.

Súbory diplomovej práce: