Offline text-independent writer identification using codebook and efficient code extraction methods

• Two new codebook based methods for writer identification are proposed.
• The introduced code extraction methods are very efficient.
• Optimal parameter values for English and Farsi handwritings are determined.
• The method is compared with the existing methods comprehensively.
• The proposed method outperforms existing methods for Farsi and English languages.

In this paper, an efficient method for text-independent writer identification using a codebook method is proposed. The method uses the occurrence histogram of the shapes in a codebook to create a feature vector for each specific manuscript. For cursive handwritings, a wide variety of different shapes exist in the connected components obtained from the handwriting. Small fragments of connected components are used to avoid complex patterns. Two efficient methods for extracting codes from contours are introduced. One method uses the actual pixel coordinates of contour fragments while the other one uses a linear piece-wise approximation using segment angles and lengths. To evaluate the methods, writer identification is conducted on two English and three Farsi handwriting databases. Both methods show promising performances with the performance of second method being better than the first one.