A Robust End-To-End Information Extraction System for Vietnamese Identity Cards

These days, automatic information extraction from identity documents is essential for many applications which require users authentication. In previous researches, this task was carried out by using image processing. However, the document identity recognition from images captured by smartphones encounters problems such as various backgrounds, perspective, blurring. Moreover, some cards may have dated backgrounds and diverse fonts printed manually by old typewriters. To solve these problems, this paper proposes an end-to-end system for extracting information from the Vietnamese identity cards captured by camera devices. Firstly, the ID card is aligned with the standard aspect ratio and orientation based on corner detection and national emblem detection. The proposed solution can locate and crop precisely the foreground target area of the oriented ID card in various contexts. In the text localization stage, a deep convolutional neural network model is used to bounding the required texts which are then classified by their positions. Finally, the text recognition is implemented based on the Attention OCR architecture. In experiments, this proposed method reduces processing time in comparison with the state of the art and the average accuracy achieved is higher 91%.