Authors : Mohammad F. Nasrudin, Elmi Mahzum, Khairuddin Omar, Choong-Yeun Liong and Mohamad Shanudin Zakaria
Abstract: This study discusses offline handwritten Jawi recognition using the trace transform. We use two attainable kinds of features from the trace transform which are the "object signature" and "triple feature". They are invariant to affine distortion, generated by the trace transform to discriminate between offline handwritten Jawi sub words. In trace transform, features construction of an image consists of tracing an image with straight lines, along which certain functional of the image function are calculated in all possible orientations. An object signature or a function of the orientation of the parallel lines is produced when a second functional is applied over all values computed along parallel lines. The computed object signature is in a form of string of numbers. A single number, triple feature is produced when a third functional over the string of numbers is applied. If the functional used have some predefined properties, the feature can be used to characterise the handwriting in an affine way. In this study, we also demonstrate the way of determining useful signatures and selecting cross-correlation methods for the signature classification. The results of the recognition experiments show that the object signature is a better feature than the triple feature in recognition of offline Jawi words.
Mohammad F. Nasrudin, Elmi Mahzum, Khairuddin Omar, Choong-Yeun Liong and Mohamad Shanudin Zakaria, 2018. Trace Transform Based Features for Offline Handwritten Jawi Word Recognition. International Journal of Soft Computing, 13: 51-60.