Legal Amount Recognition in Bank Cheques Using Capsule Networks

摘要: Legal amount detection is a decade old conundrum hindering the efficiency of automatic cheque systems. Ever since advent legal as use-case in computer vision ecosystem, it has been hampered by deficiency effective machine learning models to detect language-specific on bank cheques. Currently, convolutional neural networks are most widely used deep algorithms for image classification. Yet majority architectures fail capture information like shape, orientation, pose images due use max pooling. This paper proposes novel way extract, process and segment amounts into words from Indian cheques written English recognize them. The uses capsule cheques, which enables orientation using dynamic routing agreement techniques communication between capsules thus improves recognition accuracy.