from os import mkdir, listdir
from os.path import isdir, basename
from random import choice, randrange
from string import digits
from PIL import Image, ImageDraw # pillow
from PIL.ImageFont import truetype
from sklearn import svm
from sklearn.model_selection import train_test_split

图像尺寸、图片中的数字字体大小、噪点比例

width, height = 30, 60
fontSize = 40
noiseRate = 8

def generateDigits(dstDir='datasets', num=4000):

# 生成num个包含数字的图片文件存放于当前目录下的datasets子目录
if not isdir(dstDir):
    mkdir(dstDir)
# digits.txt用来存储每个图片对应的数字
with open(dstDir+'\\digits.txt', 'w') as fp:
    for i in range(num):
        # 随机选择一个数字，生成对应的彩色图像文件
        digit = choice(digits)
        im = Image.new('RGB', (width,height), (255,255,255))
        imDraw = ImageDraw.Draw(im)
        font = truetype('c:\\windows\\fonts\\TIMESBD.TTF',
                          fontSize)
        imDraw.text((0,0), digit, font=font, fill=(0,0,0))
        # 加入随机干扰
        for j in range(int(noiseRate*width*height)):
            w, h = randrange(1, width-1), randrange(height)
            # 水平交换两个相邻像素的颜色
            c1 = im.getpixel((w,h))
            c2 = im.getpixel((w+1,h))                
            imDraw.point((w,h), fill=c2)
            imDraw.point((w+1,h), fill=c1)
        im.save(dstDir+'\\'+str(i)+'.jpg')
        fp.write(digit+'\n')

def loadDigits(dstDir='datasets'):

# 获取所有图像文件名
digitsFile = [dstDir+'\\'+fn for fn in listdir(dstDir)
               if fn.endswith('.jpg')]
# 按编号排序
digitsFile.sort(key=lambda fn: int(basename(fn)[:-4]))
# digitsData用于存放读取的图片中数字信息
# 每个图片中所有像素值存放于digitsData中的一行数据
digitsData = []
for fn in digitsFile:
    with Image.open(fn) as im:
        data = [sum(im.getpixel((w,h)))/len(im.getpixel((w,h)))
                 for w in range(width)
                 for h in range(height)]
        digitsData.append(data)
# digitsLabel用于存放图片中数字的标准分类
with open(dstDir+'\\digits.txt') as fp:
    digitsLabel = fp.readlines()
digitsLabel = [label.strip() for label in digitsLabel]
return (digitsData, digitsLabel)

生成图片文件

generateDigits(num=4000)

加载数据

data = loadDigits()
print('数据加载完成。')

随机划分训练集和测试集，其中参数test_size用来指定测试集大小

X_train, X_test, y_train, y_test = train_test_split(data[0],

                                                data[1],
                                                test_size=0.1)

创建并训练模型

svcClassifier = svm.SVC(kernel="linear", C=1000, gamma=0.001)
svcClassifier.fit(X_train, y_train)
print('模型训练完成。')

使用测试集对模型进行评分

score = svcClassifier.score(X_test, y_test)
print('模型测试得分：', score)