驗(yàn)證碼（CAPTCHA，全自動(dòng)區(qū)分計(jì)算機(jī)和人類(lèi)的圖靈測(cè)試）的縮寫(xiě)，是一種區(qū)分用戶是計(jì)算機(jī)還是人工智能的全自動(dòng)程序。
實(shí)驗(yàn)步驟：

• 1、創(chuàng)建驗(yàn)證碼
• 2、對(duì)驗(yàn)證碼進(jìn)行01值化
• 3、降噪
• 4、對(duì)驗(yàn)證碼進(jìn)行切分
• 5、對(duì)切分后的驗(yàn)證碼進(jìn)行圖片轉(zhuǎn)數(shù)字化
• 6、使用邏輯回歸建模
• 7、對(duì)新輸入的圖片進(jìn)行預(yù)測(cè)

驗(yàn)證碼的創(chuàng)建
1、隨機(jī)生成驗(yàn)證碼的顏色
2、隨機(jī)生成驗(yàn)證碼數(shù)字
3、使用PIL進(jìn)行畫(huà)圖

import os from PIL import Image from PIL import ImageDraw from PIL import ImageFont import random import matplotlib.pyplot as plt def getRandomColor():"""獲取一個(gè)隨機(jī)顏色(r,g,b)格式的:return:"""c1 = random.randint(0, 255)c2 = random.randint(0, 255)c3 = random.randint(0, 255)if c1 == 255:c1 = 0if c2 == 255:c2 = 0if c3 == 255:c3 = 0return (c1, c2, c3) def getRandomStr():"""獲取一個(gè)隨機(jī)數(shù)字，每個(gè)數(shù)字的顏色也是隨機(jī)的:return:"""random_num = str(random.randint(0, 9))return random_num def generate_captcha():"""使用PIL畫(huà)圖步驟:return: """# 獲取一個(gè)Image對(duì)象，參數(shù)分別是RGB模式。寬150，高30， 隨機(jī)顏色image = Image.new('RGB', (150, 50), (255, 255, 255))# 獲取一個(gè)畫(huà)筆對(duì)象，將圖片對(duì)象傳過(guò)去draw = ImageDraw.Draw(image)# 獲取一個(gè)font字體對(duì)象參數(shù)是ttf的字體文件的目錄，以及字體的大小font = ImageFont.truetype("arlrdbd.ttf", size=32) # 如果找不到字體，需要從網(wǎng)上下載到本地label = "" # 隨機(jī)生成有5個(gè)數(shù)字的字符串for i in range(5):random_char = getRandomStr()label += random_char# 在圖片上寫(xiě)東西，參數(shù)是：定位，字符串，顏色，字體draw.text((10+i*30, 0), random_char, getRandomColor(), font=font)# 畫(huà)出隨機(jī)噪點(diǎn)噪線width = 150height = 30# 畫(huà)線for i in range(3):x1 = random.randint(0, width)x2 = random.randint(0, width)y1 = random.randint(0, height)y2 = random.randint(0, height)draw.line((x1, y1, x2, y2), fill=(0, 0, 0))# 畫(huà)點(diǎn)for i in range(5):draw.point([random.randint(0, width), random.randint(0, height)], fill=getRandomColor())x = random.randint(0, width)y = random.randint(0, height)draw.arc((x, y, x + 4, y + 4), 0, 90, fill=(0, 0, 0))# 保存到硬盤(pán)，名為test.png格式為png的圖片image.save(open(''.join(['captcha_images/', label, '.png']), 'wb'), 'png')# image.save(open(''.join(['captcha_predict/', label, '.png']), 'wb'), 'png')

執(zhí)行代碼之后，會(huì)在‘captcha_images’下生成實(shí)驗(yàn)所需的圖片，如圖：

圖像處理：對(duì)生成的圖片進(jìn)行處理
(1)對(duì)驗(yàn)證碼圖片二值化，首先把圖像從RGB 三通道轉(zhuǎn)化成Gray單通道，然后把灰度圖（0～255）轉(zhuǎn)化成二值圖（0，1）。
(2)將處理好的二值圖進(jìn)行降噪，去除圖片中的噪點(diǎn)和噪線

from PIL import Image import numpy as np import matplotlib.pyplot as plt import osdef binarization(path):"""把一個(gè)rgb的圖轉(zhuǎn)換成一個(gè)二值圖:param path::return:"""# 通過(guò)path把圖像laod進(jìn)來(lái)img = Image.open(path)# 把圖像轉(zhuǎn)化成一個(gè)灰度圖img_gray = img.convert("L")# 把灰度圖組裝成數(shù)組形式img_gray = np.array(img_gray)# print(img_gray)# 得到灰度圖的寬和高w, h = img_gray.shapefor x in range(w):for y in range(h):# 得到每一個(gè)像素塊里的灰度值gray = img_gray[x, y]# 如果灰度值小于等于220， 就把它變成黑色if gray <= 220:img_gray[x, y] = 0# 如果灰度值大于220，就把它變成白色else:img_gray[x, y] = 1plt.figure("")plt.imshow(img_gray, cmap="gray")plt.axis("off")plt.show()return img_graydef noiseReduction(img_gray, label):"""降噪，也就是處理離群點(diǎn)如果一個(gè)像素點(diǎn)周?chē)挥行∮?個(gè)黑點(diǎn)的時(shí)候，那么這個(gè)點(diǎn)就是離群點(diǎn):param img_gray::param label::return:"""height, width = img_gray.shapefor x in range(height):for y in range(width):cnt = 0# 白色的點(diǎn)不用管if img_gray[x, y] == 1:continueelse:try:if img_gray[x-1, y-1] == 0:cnt += 1except:passtry:if img_gray[x-1, y] == 0:cnt += 1except:passtry:if img_gray[x-1, y+1] == 0:cnt += 1except:passtry:if img_gray[x, y-1] == 0:cnt += 1except:passtry:if img_gray[x, y+1] == 0:cnt += 1except:passtry:if img_gray[x+1, y-1] == 0:cnt += 1except:passtry:if img_gray[x+1, y] == 0:cnt += 1except:passtry:if img_gray[x+1, y+1] == 0:cnt += 1except:passif cnt < 4: # 周?chē)儆?個(gè)點(diǎn)就算是噪點(diǎn)img_gray[x, y] = 1plt.figure(" ")plt.imshow(img_gray, cmap="gray")plt.axis("off")plt.savefig("".join(["clean_captcha_img/", label, ".png"]))def image_2_clean():"""把所有的圖像都轉(zhuǎn)化成二值圖:return:"""captchas = os.listdir("".join(["captcha_images/"]))for captcha in captchas:label = captcha.split(".")[0]image_path = "".join(["captcha_images/", captcha])# 二值化im = binarization(image_path)# 降噪noiseReduction(im, label)if __name__ == '__main__':image_2_clean()# path = "captcha_images/00006.png"# img_gray = binarization(path)# noiseReduction(img_gray, label='00006')

圖像分割：對(duì)降噪后的圖片進(jìn)行分割，并對(duì)分割后的圖片進(jìn)行存儲(chǔ)

import os from PIL import Image from PIL import ImageDraw from PIL import ImageFont import random import matplotlib.pyplot as plt def cutImg(label):"""把圖像的每一個(gè)數(shù)字都切分出來(lái)，并且存到新的文件夾下:param label::return:"""labels = list(label)img = Image.open("".join(['clean_captcha_img/', label, '.png']))for i in range(5):pic = img.crop((100*(1+i), 170, 100*(1+i)+100, 280))plt.imshow(pic)# seq就是我們需要存到文件的文件名seq = get_save_seq(label[i])pic.save("".join(["cut_number/", str(label[i]), "/", str(seq), '.png'])) def get_save_seq(num):"""得到需要保存的數(shù)據(jù)的文件名每一個(gè)數(shù)文件下的文件名，都是從0開(kāi)始保存 0.png, 1.png....:param num::return:"""nmlist = os.listdir("".join(["cut_number/", num, "/"]))if len(nmlist) == 0 or nmlist is None:return 0else:max_file = 0for file in nmlist:if int(file.split(".")[0]) > max_file:max_file = int(file.split(".")[0])return int(max_file) + 1 def clean_to_cut():"""對(duì)每一個(gè)文件都進(jìn)行切分:return:"""captchas = os.listdir("".join(["clean_captcha_img"]))for captcha in captchas:label = captcha.split(".")[0]cutImg(label) def create_dir():for i in range(10):os.mkdir("".join(["cut_number/", str(i)]))if __name__ == '__main__':# create_dir()clean_to_cut()

圖片轉(zhuǎn)數(shù)字化：對(duì)切分后的圖片灰度化、二值化，使用Image.open()打開(kāi)圖片文件，得到plt圖片對(duì)象，將plt圖片對(duì)象轉(zhuǎn)換為ndarray對(duì)象，將二值化后的圖像轉(zhuǎn)化為1行n列，存入X列表中，并將其對(duì)應(yīng)的數(shù)字存入Y列表中。
模型的生成：將X，Y傳入邏輯回歸模型中，使用交叉驗(yàn)證和網(wǎng)格搜索尋找最優(yōu)的參數(shù)。

import os from PIL import Image import numpy as np import matplotlib.pyplot as plt from sklearn.linear_model import LogisticRegression from sklearn.model_selection import train_test_split from sklearn.model_selection import GridSearchCV from sklearn.externals import joblibfrom sklearn.metrics import confusion_matrix from sklearn.metrics import precision_score from sklearn.metrics import recall_score def load_data():"""把數(shù)據(jù)從cut_number里面導(dǎo)出來(lái)其中X指的是每一個(gè)數(shù)字的01值的排列， Y指的是這個(gè)數(shù)字是什么:return:"""X, Y = [], []cut_list = os.listdir("cut_number")# 循環(huán)cut_number文件夾下的每一個(gè)自文件夾（1，2，3，4，5...）for numC in cut_list:num_list_dir = "".join(["cut_number/", str(numC), "/"])nums_dir = os.listdir(num_list_dir)# 循環(huán)子文件夾中的每一個(gè)圖片# print(np.array(Image.open(''.join(['cut_number/', str(numC), '/', '0.png']))))for num_file in nums_dir:# 導(dǎo)入數(shù)字圖片img = Image.open("".join(["cut_number/", str(numC), "/", num_file]))# print(np.array(img))# 對(duì)數(shù)字圖片做灰度化img_gray = img.convert("L")# plt.imshow(img_gray)# 把灰度化圖片保存到數(shù)組里img_array = np.array(img_gray)w, h = img_array.shape# 把灰度化的圖片做二值化for x in range(w):for y in range(h):gray = img_array[x, y]if gray <= 220:img_array[x, y] = 0else:img_array[x, y] = 1# 把二值化的圖片reshape成1行，n列img_re = img_array.reshape(1, -1)# print(img_re[0])X.append(img_re[0])Y.append(int(numC))return np.array(X), np.array(Y)def generate_model(X, Y):"""生成模型:param X::param Y::return:"""# 區(qū)分測(cè)試集和訓(xùn)練集，37開(kāi)X_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=0.3)log_clf = LogisticRegression(multi_class="ovr", solver="sag", max_iter=10000)# log_clf.fit(X_train, Y_train)# 利用交叉驗(yàn)證選擇參數(shù)param_grid = {"tol": [1e-4, 1e-5, 1e-2], "C": [0.4, 0.6, 0.8]}grid_search = GridSearchCV(log_clf, param_grid=param_grid, cv=3)grid_search.fit(X, Y)print(grid_search.best_params_)print("模型生成成功")# 將模型持久化joblib.dump(log_clf, "captcha_model/captcha_model.model")print("模型保存成功")if __name__ == '__main__':X, Y = load_data()generate_model(X, Y)

圖片的預(yù)測(cè)：
輸入要預(yù)測(cè)的圖片，對(duì)其進(jìn)行灰度化，二值化，并進(jìn)行分割，將分割出來(lái)的五個(gè)圖片輸入進(jìn)模型中。

import os from PIL import Image import numpy as np import matplotlib.pyplot as plt from sklearn.linear_model import LogisticRegression from sklearn.model_selection import train_test_split from sklearn.model_selection import GridSearchCV from sklearn.externals import joblibfrom sklearn.metrics import confusion_matrix from sklearn.metrics import precision_score from sklearn.metrics import recall_score from .captcha_logistic import *def get_model():model = joblib.load('captcha_model/captcha_model.model')return modeldef model_predict():path = 'captcha_predict/unknown.png'pre_img_gray = binarization(path)noiseReduction(pre_img_gray, 'unknown')# cut imagelabels = ['0', '1', '2', '3', '4']img = Image.open(''.join(['clean_captcha_img/unknown.png']))for i in range(5):pic = img.crop((100 * (1 + i), 170, 100 * (1 + i) + 100, 280))plt.imshow(pic)pic.save(''.join(['captcha_predict/', labels[i], '.png']))result = ''model = get_model()for i in range(5):path = ''.join(['captcha_predict/', labels[i], '.png'])img = Image.open(path)img_gray = img.convert('L')img_array = np.array(img_gray)w, h = img_array.shapefor x in range(w):for y in range(h):gray = img_array[x, y]if gray <= 220:img_array[x, y] = 0else:img_array[x, y] = 1img_re = img_array.reshape(1, -1)X = img_re[0]y_pre = model.predict([X])result = ''.join([result, str(y_pre[0])])return result if __name__ == '__main__':result = model_predict()print(result)

總結(jié)

以上是生活随笔為你收集整理的机器学习案例：验证码识别(Captcha)的全部?jī)?nèi)容，希望文章能夠幫你解決所遇到的問(wèn)題。

如果覺(jué)得生活随笔網(wǎng)站內(nèi)容還不錯(cuò)，歡迎將生活随笔推薦給好友。