Online Urdu Handwriting Recognition System Using Geometric Invariant Features
Z. Jan, Muhammad Shabir, Mohammad A. U. Khan, A. Ali, Muhammad Muzammal · The Nucleus · 2016
Online touch sensitive devices facilitate users by providing an easy way for inputting online handwritten text. Many useful applications are developed for other cursive script language and are practically used in different fields like banking, commerce, academics, administration and education etc. There are also some systems proposed for online Urdu handwriting recognition but either they have low accuracy rates or high constraints on user while writing. Online Urdu handwriting recognition is a difficult task due to its cursive property and writing complexity. The proposed system tries to recognize Urdu characters and words by using geometric features i.e. cosine angles of trajectory, discrete fourier transform of trajectory, inflection points, self-intersections, convex hull, radial feature, grid (orthogonal and perspective), and retina feature. The proposed system is font, rotation, scale and shift invariant due to geometric invariant features. Before feature extraction low pass filtering and resampling is applied on each input stroke trajectory to remove noise caused by input device and hand movement. After feature extraction linear support vector machine is used for training and testing which gives up to 97% classification accuracy on test data. In recognition phase the proposed system gives a very low false rejection rate.