Geometrical and Structural Features for Forensics in Handwritten Bank Cheques
Anupam Wadhwa, Mohak Maheshwari, Prabhat Dansena, Soumen Bag · 2018
In handwritten documents like bank cheques, adding new words can lead to financial loss. Moreover, identification of such words becomes more challenging when number of added words are very less. Since these documents are processed in digital form on daily basis, problem becomes even more worse to locate and to identify such changes in digital copy of these documents. In this regard, we propose a classification based approach that aims to identify whether pair of words are written by the same person or not. For this purpose, we use geometrical and structural features like stroke width, direction, inter character space, and so on. We use five classifiers, namely kNN, Decision tree, Radial basis-SVM, Multi-layer Perceptron (MLP), and Random Forest to train the model. The numerical results reveal that Random Forest classifier outperforms the other classifiers. The proposed scheme performs efficiently on IAM and IDRBT data sets using Random Forest classifier with average accuracy of 83.92% and 70.93% respectively.