Certificate information detection and extraction method

Abstract

The invention discloses a certificate information detection and extraction method. A synthesized certificate data set is used to replace a real certificate data set for the training of a Faster-RCNN detection model; certificate information is detected and classified by using a method based on a deep neural network, and special preprocessing is not needed; and an end-to-end character recognition method is used for recognizing the certificate character information with the uncertain length, and segmentation is not needed. The character length does not need to be set, a single Chinese character does not need to be recognized, the influence on the recognition rate due to segmentation errors is avoided, and the text information can be obtained only by inputting a picture into the network model.Compared with a traditional method based on character template matching, the method has strong adaptability when facing the problems of low brightness, low contrast, uneven illumination, deformation,incompleteness, shielding and the like, and the precision of the method is far superior to that of the traditional method.

Description

technical field [0001] The invention relates to a certificate information detection and extraction method, which belongs to the technical fields of image processing and character recognition. Background technique [0002] Deep neural network is a complex mathematical model. It is a mode of deep learning. It consists of layers of network layers. The input data gets output data after passing through all network layers. According to the difference between the output data and the labeled data, it can be Constructing a loss function, and then backpropagating the gradient of the loss function, the weights of the network layer can be updated, thereby further reducing the difference between the output data and the labeled data. Among them, the data set used for input and the corresponding labeled data constitute the training data set of the deep neural network, and the function and performance of the deep neural network are related to the network structure and the training data set....

AI Technical Summary

Problems solved by technology

Relying on rules for text classification has two disadvantages: (1) it does not have generalization performance, and a set of rules needs to be designed for each scene; (2) the rules are either extremely complex or poor in robustness, and require high image quality

Images