Smart Phone Aided Intelligent Invoice Reimbursement System

Abstract

Invoice reimbursement is one of indispensable aspects of business in many countries especially in China. Conventional manpower based reimbursement schemes often lead to high cost and inefficiency and robot based reimbursement systems require large space and huge equipment costs. In order to solve these problems, we propose an smart phone aided reimbursement system to realize the intelligent localization and identification in invoice images. First, invoice image is taken by camera of smart phone. Second, the Hough transform is used to detect the linear principle to correct the tilt of the invoice image with different background and different tilt angles. Third, we adopt You Only Look Once-Version 3 (YOLOv3) based target detection network to train the tagged data set, to obtain the training weights, and then realize the intelligent positioning and extraction. Finally, the invoice information is identified using optical character recognition (OCR). Experiment results are given to verify that the localization accuracy can reach 92.5% when the intersection over union (IoU) is set as 0.5 and the identification accuracy can reach up to 97.5% for invoice information.