Verification Code Recognition

This repository is mainly about auto recognizing verification code. A method called "Drip algorithm" is used to cut the picture into parts which is convenient for subsequent identification operations.

Introduction

./cut_algorithm: Mainly cut algorithm programed in matlab language.

./cut_algorithm/leach.m: Function that used to leach noise dot.

./cut_algorithm/smooth.m: Function that used to smooth the picture.

./cut_algorithm/data2_type8.m: Main script to run which is mainly focused on the deta in ./data/4nngn1. ./data: Several types of captcha data, which is named by some principles:

Folder name meaning analysis

First marker character: 4 5 6

1. 4 There are four characters in the picture

2. 5 There are five characters in the picture

3. 6 There are six characters in the picture

Second mark character: s n

1. s single color (the color of the character to be recognized, regardless of the background color)

2. n color, non-single color

Third marker character: x n

1. x characters have rotation and overlap, scattered arrangement

2. n characters do not have rotation and overlap, neatly arranged

Fourth marker character: g n

1. g There is interference (referring to the long horizontal line of the foreground,excluding the short lines and noise points of the background, etc.)

2. n no interference

Fifth marker character: q n

1. q character is distorted
2. n no distortion

The sixth and subsequent mark characters: A a 123

1. A means uppercase letters are included
2. a means that lowercase letters are included
3. 123 means contains numbers

./data_cut: Algorithm cut result, each character is saved in its belonging folder. ./source: Tensorflow is used to recognize the cutted subimage. ./captcha_data: Train and Test data for CNN.

Source images

Cut subimage samples

B cut result

N cut result

Contact

Email: shp395210@outlook.com