Artificial synthesizing method of training samples and verification code recognizing method based on samples

Abstract

The invention discloses an artificial synthesizing method of training samples and a verification code recognizing method based on the samples. The artificial synthesizing method of the training samples comprises the following steps: S1, generating a verification code lexicon; S2, generating a background photo gallery; S3, screening rectangular background blocks; S4, collecting typeface sets; S5, randomly matching words and expressions with the background blocks; and S6, taking the background blocks which have been written with the words and expressions as samples. The invention further discloses a verification code recognizing method based on the training samples. The verification code recognizing method based on the training samples comprises the following steps: S1, feature extraction; S2, sequence calibration; and S3, correction of recognition results. The training samples are acquired by a mode of synthesizing samples artificially, only recognized character types need to be set, but samples in the corresponding types can be generated, good effect can be achieved by adding a small amount of actual website samples, and a large number of labors and financial resources for sample collection are reduced while the samples are diversified.

Description

technical field [0001] The invention relates to the technical field of the Internet, in particular to a method for artificially synthesizing a training sample and a verification code recognition method based on the sample. Background technique [0002] In general, image-based text recognition includes optical character recognition (Optical Character Recognition, OCR) based on scanned text and CAPTCHA (Completely Automated Public Turing Test to Tell Computers and Humans Apart, which is widely used in website registration verification, fully automatic distinction between computers and humans. Turing test). In comparison, scanner-based OCR is the easiest, and CAPTCHA is the hardest. [0003] Traditional verification code cracking mainly detects characters, cuts them, and finally recognizes individual characters. This method can only recognize traditional verification codes with clear backgrounds and relatively correct and simple characters, but cannot adapt to verification co...

AI Technical Summary

Problems solved by technology

This method can only recognize traditional verification codes with clear backgrounds and relatively correct and simple text, but cannot adapt to verification codes with complex backgrounds.

At present, there is also a verification code recognition method of cnn+rnn machine learning, but it is relatively simple to synthesize training samples through a verification code generator, and the workload of manual collection of samples is heavy, so it cannot be implemented in actual projects.

Images