Effective feature descriptor-based new framework for off-line text-independent writer identification

Feature engineering is a key factor of machine learning applications. It is a fundamental process in writer identification of handwriting, which is an active and challenging field of research for many years. We propose a conceptually computationally efficient, yet simple and fast local descriptor referred to as Block Wise Local Binary Count (BW-LBC) for offline text-independent writer identification of handwritten documents. Proposed BW-LBC operator, which characterizes the writing style of each writer, is applied to a set of connected components extracted and cropped from scanned handwriting samples (documents or set of words/text lines) where each labeled component is seen as a texture image. The feature vectors computed from the components in all the writing samples are then fed to the 1NN (Nearest Neighbor) classifier to identify the writer of the query documents. Simulated experiments are performed on three challenging and publicly available handwritten databases (IFN/ENIT, AHTID/MW, and CVL) containing handwritten texts in Arabic and English languages, respectively. Experimental results show that our proposed system combined with BW-LBC descriptor demonstrate superior performance on the Arabic script and competitive performance on the English one against the old and recent writer identification systems of the state-of-the-art.