A Text Detection and Recognition Method for Receipt Image

Abstract

The invention discloses a bill image text detection and recognition method based on deep learning. First, a convolutional neural network is used to perform feature extraction on the bill image to generate a first feature map. Then the text detection network performs multi-task prediction through classification and regression operations on the first feature map to obtain the detection text box. On this basis, the text recognition network maps the detected text box to the corresponding area of ​​the first feature map and performs a pooling operation to generate a second feature map with a fixed height and a proportional change in width, and convert the second feature map into a feature sequence. The context information of the feature sequence is encoded by a recurrent neural network, and then decoded by a set of recurrent neural networks with attention mechanism to obtain the recognition result of the text region. The detection task and the recognition task are integrated into a unified network framework, which not only realizes the feature sharing of the convolutional layer, but also completes the end-to-end joint training to improve the overall recognition performance of the model.

Description

technical field [0001] The invention belongs to the technical field of image text detection and recognition, and more specifically relates to a method for text detection and recognition of bill images. Background technique [0002] Financial bills are one of the important vouchers in the circulation process of the national financial market. They undertake the capital circulation business among social entities such as individuals, enterprises, and banks. Common types of bills include checks, bills of exchange, and certificates of deposit. With the rapid development of our country's economy, all kinds of bill business have also developed rapidly, and bill voucher processing is an important task in the daily business of banks. Nowadays, in cash withdrawal and other business fields, automatic or semi-automatic processing has basically been realized. For example, various banknote counters, sorting machines, counterfeit detectors and self-service deposit and withdrawal terminals o...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the defects of the prior art, the purpose of the present invention is to provide a method for text detection and recognition of bill images based on deep learning, which avoids the multi-step processing flow of traditional OCR technology and the problem of low text recognition accuracy in complex scenes , to ensure real-time processing while improving detection and recognition performance

Images