Overlapping Handwritten Text Identification Based Forgery Detection in Bank Cheques using PHLS and SGCIP-CNN Techniques

Abstract Segmentation of handwritten Bank Cheque Images (BCI) aids in extracting significant details and makes smoother transactions. Nevertheless, overlapped Handwritten Texts (HT) with Printed Texts (PT) were not focused in previous works, thereby disrupting the segmentation process. Thus, Intelligent Character Recognition (ICR) and Pseudo Letter with Height-based Segmentation (PLHS) are utilized in this framework to efficiently remove the overlapped texts for accurate HT segmentation. Initially, for obtaining a denoised enhanced image, the BCI is preprocessed. The overlapped HT with PTs are recognized and segmented separately from the enhanced BCI using ICR and PHLS, respectively. After that, by utilizing Nanonets, the date, signature, name, and amount from HT are separately identified. Then, by using texture inpainting, the faded texts are redrawn followed by edge detection and contour formation. Lastly, by utilizing a Sigmoidal Growing Cosine Intermap Pooling-based Convolutional Neural Network (SGCIP-CNN), the recognized texts are categorized as forged and genuine cheques with a classification accuracy of 98.7899% for effective money transactions. Hence, the proposed work performed better than the prevailing techniques.