Method for identifying abnormal signature and system thereof

Abstract

The invention discloses a method for identifying an abnormal signature. The method comprises the following steps: preprocessing a sample signature picture and forming a signature sample library; constructing a convolutional neural network model, setting an input layer, a convolutional layer, a pooling layer, a full connection layer and a classification layer, and setting convolutional neural network model parameters; optimizing the convolutional neural network model; and training a convolutional neural network model by using the training set sample, and testing the accuracy and stability of the trained convolutional neural network by using the test set; acquiring a client signature image and preprocessing the client signature image, and converting the image into a matrix with the same size; and calling an output classification model of the optimized convolutional neural network model to judge the preprocessed customer signature image data, determining an abnormal signature, and addingthe signature image data into a signature sample library to perform model continuous training. According to the invention, the computer identification technology is innovatively applied to the identification of non-standard signatures, and the output effect is ideal in combination with the deep learning mode.

Description

technical field [0001] The invention relates to the technical field of signature identification, in particular to a method and system for identifying abnormal signatures. Background technique [0002] The user's signature represents the user's subjective confirmation behavior, which is widely used in various business links and has extremely important significance. However, currently, in the entire process of obtaining user signatures and retaining files, the identification of user signature content still remains at the level of manual identification and simple data index screening. The former requires a lot of human resources, while the latter is not enough in terms of accuracy. As expected. And the present invention mainly identifies abnormal signatures in the user's signature process, such as drawing circles, drawing dots and drawing lines, and so on. Machine learning is used to achieve high-accuracy, highly automated and intelligent signature recognition. For scenarios ...

AI Technical Summary

Problems solved by technology

[0012] The manual review method is unrealistic when transactions are frequent and will reduce the efficiency of transactions

It is also not feasible for business parties such as merchants to request to monitor user signature behavior, because for merchants, it is the intrinsic motivation of merchants to allow users to complete transactions as soon as possible and increase transaction volume, and they will not specifically regulate user signature behaviors, because The probability of chargeback and order adjustment is relatively low

Calculate the proportion of pixels and set the threshold, the misjudgment rate is very high, for example, if the user name is relatively simple, such as "Ding Yi", it will be misjudged

In addition, this method is easy to be cracked, such as graffiti to increase the proportion of pixels, which is also an abnormal signature but the system cannot judge normally

Images