PREPROCESSING AND SEGMENTATION TECHNIQUE FOR AN OFFLINE NOVEL GURMUKHI HANDWRITTEN DATASET

Abstract

Writer identification systems are advanced technology solutions used in forensic investigation, authentication of historical document and the identification of writer that are crucial for identifying handwritten text with specific individuals. This study presents a novel dataset called GUR-PRIMTA of 2100 handwritten text images in the Gurmukhi script which is often used in Punjab, India. The experimental setup involved many crucial phases, primarily segmentation and preprocessing. Using deep convolutional neural networks (CNNs) to reduce and eliminate noise from images, VGG16 model is used to preprocess the handwritten text images. A contour curving-based segmentation approach has been applied to segment handwritten text words for further analysis. The proposed methodology shows effectiveness to enhance the accuracy and speed in the tasks of writer identification, with implications for various real-world applications that required accurate features of handwritten text.