Arabic Writer Identification System Using the Histogram of Oriented Gradients (HOG) of Handwritten Fragments
Yaâcoub Hannad, Imran Siddiqi, Youssef El Merabet, Mohamed El Youssfi El Kettani · 2016
This paper 1 presents an enhanced approach for writer identification from offline Arabic handwriting samples in text-independent mode. Based on the hypothesis that graphical fragments in handwriting are individual, we propose a technique based on texture analysis where the handwriting is divided into small fragments and each fragment is represented by the histogram of oriented gradients (HOG). The set of HOG descriptors for all the fragments in the writing is used to characterize its writer. The proposed system is evaluated using writing samples of the IFN/ENIT database realizing an identification rate of 86.62% on 411 writers.