Comparison of global and local features for author’s identification by using geometrical and zoning methods

Identification analysis for author's handwriting image in forensic investigation is still an important research area in this current big data era. Images feature extraction can lead to an issue of high dimensionality of data. The process of feature extraction is the most crucial process in author's identification. It is important to choose the best method to represent the image. This study compared two feature extraction methods, namely Higher-Order United Moment Invariant (HUMI) and the Edge-based Directional (ED) method that construct the Global and Local Features respectively. The additional process of discretization was implemented before the training and testing phase to represent the generalized features for the classifier models. This process induced a better performance accuracy for both methods where the discretized Local Features achieved 99.95% accuracy rate that slightly outperforms the discretized Global Features with only 99.91%.