Smart text extraction system for Bank Cheque Images using DWT and dynamic thresholding
Neha Thakur, Deepika Ghai, Sunpreet Kaur Nanda, Sandeep Kumar, Mandeep Kaur · 2025
Bank cheques are largely utilized for financial dealings or transactions, with millions processed daily worldwide. A significant challenge in cheque management is the high cost and time involved in processing, which could be mitigated by automating the cheque processing system. While considerable research has focused on extracting certain fields—such as the date, signature, and legal and courtesy amounts—less attention has been given to fields like the bank logo, bank name, and payee’s name. This chapter presents a novel and effective approach for automatically extracting various data fields from bank cheque images, aiming for improved precision, recall, and reduced processing time. This chapter introduces a hybrid technique that integrates Discrete Wavelet Transform (DWT) with dynamic thresholding and a logical AND operator for data extraction. Text features exhibit abrupt variations and distinct edges when transformed using wavelets. Initially, edges are identified in the input grayscale image through 2D DWT, resulting in detailed sub-bands that include both text and non-text regions. Next, utilizing dynamic thresholding, morphological dilation techniques are used to join individual text regions inside these sub-bands. Finally, the logical AND operator and area-based filtering techniques are employed to precisely identify the data fields within the bank cheque images. The proposed technique outperforms the currently available methods in terms of Precision Rate (PR), Recall Rate (RR), and Processing Time (PT) for obtaining relevant data fields under varied circumstances, according to MATLAB experiments conducted on both public and own datasets.