数字加减法验证码 识别 使用K-近邻算法(KNN)

Posted 2020-12-16 dockers

验证码如上所示

100*30

下面咱们开始神奇的旅程

 下载批量验证码图片数据集用来训练   

此验证码比较简单就下载了500

#!/usr/bin/env python
# -*- coding:utf-8 -*-

# Author: zhibo.wang
# E-mail: [e-mail]
# Date  : 20/06/29 10:06:45
# Desc  :


import random
import requests
import uuid
import time
import os

def is_exists(path_):
    if not os.path.exists(path_):
        os.makedirs(path_)


is_exists("source/")

for i in range(0, 500):
    print(i)
    url = ‘https://www.okcis.cn/php/checkUser/code.php‘
    resp = requests.get(url, headers={
        "user-agent": "Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (Khtml, like Gecko) Ubuntu Chromium/75.0.3770.90 Chrome/75.0.3770.90 Safari/537.36"})
    with open(‘source/‘ + str(uuid.uuid4()) + ‘.png‘, ‘wb‘) as f:
        f.write(resp.content)
    time.sleep(0.1)

 二值化并切割验证码

#!/usr/bin/env python
# -*- coding:utf-8 -*-

# Author: zhibo.wang
# E-mail: [e-mail]
# Date  : 20/06/29 10:06:45
# Desc  :

import os
import uuid
import numpy as np
from PIL import Image
import time


def is_exists(path_):
    if not os.path.exists(path_):
        os.makedirs(path_)


is_exists("letters/")


def splitAndSave(path):
    path = ‘./source/‘ + path
    pix = np.array(Image.open(path).convert(‘L‘))
    pix = (pix > 200) * 255 
    split_parts = [
        [3, 23],
        [23, 43],
        [43, 63]
    ]

    for part in split_parts:
        letter = pix[0:, part[0]: part[1]]
        im = Image.fromarray(np.uint8(letter))
        save_path = ‘./letters/‘ + str(uuid.uuid4()) + ‘.png‘
        print(‘	‘, save_path)
        im.save(save_path)


if __name__ == ‘__main__‘:
    im_paths = filter(lambda fn: os.path.splitext(fn)[1].lower() == ‘.png‘,
    os.listdir(‘./source‘))
    for im_path in im_paths:
        print(im_path)
        splitAndSave(im_path)

给切割好的数据打标签

每次选中一个类型的数据放入复制粘贴到train 文件夹下  然后 修改 n 字段进行每个类别的自动修改

#!/usr/bin/env python
# -*- coding:utf-8 -*-

# Author: zhibo.wang
# E-mail: [e-mail]
# Date  : 20/06/29 11:36:11
# Desc  :

import os

def list_all_files(rootdir):
    _files = []
    try:
        list = os.listdir(rootdir)
        for i in range(0, len(list)):
            path = os.path.join(rootdir, list[i])
            if os.path.isdir(path):
                _files.extend(list_all_files(path))
            if os.path.isfile(path):
                _files.append(path)
    except Exception:
        pass
    return _files


def rename_filename(filename, newfilename):
    os.rename(filename, newfilename)



def run():
    n = "10"
    path_ = ‘train‘
    all_files = [i for i in list_all_files(path_) if len(i.split("/")[-1]) == 40]
    for i in range(0, len(all_files)):
        file_name = all_files[i]
        new_file_name = "{0}/{1}:{2}.png".format(path_, n, i)   # win文件名不能有:请自行修改
        print(new_file_name)
        rename_filename(file_name, new_file_name)


if __name__ == "__main__":
    run()

数据打标签完成开始训练模型

训练数据 knn

 

#!/usr/bin/env python
# -*- coding:utf-8 -*-

# Author: zhibo.wang
# E-mail: [e-mail]
# Date  : 20/06/29 10:06:45
# Desc  :

import os
from PIL import Image
import numpy as np
import joblib
from sklearn.neighbors import KNeighborsClassifier
from utils import list_all_files


def list_all_files(rootdir):
    _files = []
    try:
        list = os.listdir(rootdir)
        for i in range(0, len(list)):
            path = os.path.join(rootdir, list[i])
            if os.path.isdir(path):
                _files.extend(list_all_files(path))
            if os.path.isfile(path):
                _files.append(path)
    except Exception:
        pass
    return _files



def load_dataset():
    X = []
    y = []
    path_ = "train"
    all_flies = list_all_files(path_)
    keys = [str(i) for i in range(1, 11)] + ["+", "-"]
    for k in keys:
        for file_ in all_flies:
            if file_.split("/")[-1].split(":")[0] == str(k):
                path = file_
                pix = np.asarray(Image.open(path).convert("L"))
                X.append(pix.reshape(20*30))
                y.append(k)
    return np.asarray(X), np.asarray(y)


if __name__ == "__main__":
    X, y = load_dataset()
    knn = KNeighborsClassifier()
    knn.fit(X, y)
    joblib.dump(knn, ‘knn.pkl‘)

训练完成下来测试下效果

#!/usr/bin/env python
# -*- coding:utf-8 -*-

# Author: zhibo.wang
# E-mail: [e-mail]
# Date  : 20/06/29 10:06:45
# Desc  :

import os
import numpy as np

from PIL import Image
import joblib


def split_letters(path):
    pix = np.array(Image.open(path).convert(‘L‘))
    pix = (pix > 200) * 255

    split_parts = [
        [3, 23],
        [23, 43],
        [43, 63]
    ]
    letters = []
    for part in split_parts:
        letter = pix[0:, part[0]: part[1]]
        letters.append(letter.reshape(20*30))
    return letters


def get_captcha_result(model_path, filename):
    sipo_knn = joblib.load(model_path)
    letters = split_letters(filename)
    return "".join([str(i) for i in sipo_knn.predict(letters)])


if __name__ == "__main__":
    for test in os.listdir(‘./test‘):
        datas = test, get_captcha_result(‘knn.pkl‘, ‘./test/‘ + test)
        print(datas)

 

结果挺满意的百分百识别正确

下来就是进行简单的计算了   此处略。。。。。。。。。