如何在 Python 中创建带有英文字母（AZ、az、0-9）的图像？

Question

I want to prepare a dataset for the recognition of English Alphanumeric characters.我想准备一个用于识别英文字母数字字符的数据集。 I want to train my Deep learning model to recognize the characters.我想训练我的深度学习 model 来识别字符。 Image size should be 32 * 32 or 28 * 28. And Characters should be scaled with the height and width of the Image.图像大小应为 32 * 32 或 28 * 28。字符应随图像的高度和宽度缩放。 Given below is the expected Output.下面给出的是预期的 Output。

Character "A" scaled with the Image用图像缩放的字符“A”

I have tried getsize() but I am not able to scale character with the image size.我试过 getsize() 但我无法用图像大小缩放字符。

from PIL import Image, ImageDraw, ImageFont
import os,glob
import numpy as np
import random

parentPath=r"CaptchaDataset\\"


charactersList=['a','b','c','d','e','f','g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v','w','x','y','z','A','B','C','D','E','F','G','H','I','J','K','L','M','N','O','P','Q','R','S','T','U','V','W','X','Y','Z','0','1','2','3','4','5','6','7','8','9']
fontTypes=[]

for font in glob.glob("fonts\\*"): # Create Image for every font
    fontTypes.append(os.path.abspath(font))
    
for index,character in enumerate(charactersList):
    
    path=os.path.join(parentPath,character+"_"+str(index))
    os.mkdir(path)
    
    
    for imageCounter in range(len(fontTypes)):
        for repeats in range(3): #Number of Images
            img = Image.new('1', (28, 28), color = 'black')
            fnt = ImageFont.truetype(fontTypes[imageCounter], random.randint(30,50))
            w,h=fnt.getsize(character)
            d = ImageDraw.Draw(img)
            d.text(((28-w)/2-5,((28-h)/2)-5), character, font=fnt, fill=(255),align="center") #TO ALIGN CHARACTER IN CENTER
            img=np.pad(img,pad_width=10, mode='constant', constant_values=0) #Manually added padding
            img=Image.fromarray(img)
            img.save(path+"\\"+str(imageCounter)+"_"+str(repeats)+".jpg")

UnExpected Output of Above program.上述程序的意外 Output。 "g" is cropped “g”被裁剪

Answer 1

The solution is simple: Create the image after creating the font.解决方案很简单：创建字体后创建图像。

Instead of creating 28x28 pixels image, and drawing a character that may be larger than the image, create the font, get the size, and create the image according to the font's size.与其创建 28x28 像素的图像，并绘制一个可能比图像大的字符，不如创建字体，获取大小，然后根据字体大小创建图像。

Create the font, and get the font's size:创建字体，并获取字体的大小：

 fnt = ImageFont.truetype(fontTypes[imageCounter], random.randint(30, 50)) w, h = fnt.getsize(character)

Set image size according to font size and create the image:根据字体大小设置图像大小并创建图像：

 img_w, img_h = w + 20, h + 20 # Add 20 pixels padding (assume 10 pixels from each side). img = Image.new('L', (img_w, img_h), color='black') # Replace '1' with 'L' (8-bit pixels, black and white - we fill 255 so we can't use 1 bit per pixel)

Draw the character at the center of the image:在图像的中心绘制角色：

 d.text(((img_w-w)/2, (img_h-h)/2), character, font=fnt, fill=255, align="center") # TO ALIGN CHARACTER IN CENTER

Here is a complete (updated) code:这是一个完整的（更新的）代码：

from PIL import Image, ImageDraw, ImageFont
import os,glob
import numpy as np
import random

parentPath = "CaptchaDataset\\"

charactersList=['a','b','c','d','e','f','g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v','w','x','y','z','A','B','C','D','E','F','G','H','I','J','K','L','M','N','O','P','Q','R','S','T','U','V','W','X','Y','Z','0','1','2','3','4','5','6','7','8','9']
fontTypes=[]

for font in glob.glob("fonts\\*"): # Create Image for every font
    fontTypes.append(os.path.abspath(font))
    
for index,character in enumerate(charactersList):    
    path = os.path.join(parentPath,character + "_" + str(index))

    if not os.path.exists(path):
        os.mkdir(path)
     
    for imageCounter in range(len(fontTypes)):
        for repeats in range(3): # Number of Images
            #img = Image.new('1', (28, 28), color = 'black')
            fnt = ImageFont.truetype(fontTypes[imageCounter], random.randint(30, 50))
            w, h = fnt.getsize(character)
            img_w, img_h = w + 20, h + 20  # Add 20 pixels padding (assume 10 pixels from each side).
            img = Image.new('L', (img_w, img_h), color='black')  # Replace '1' with 'L' (8-bit pixels, black and white - we fill 255 so we can't use 1 bit per pixel)
            d = ImageDraw.Draw(img)
            d.text(((img_w-w)/2, (img_h-h)/2), character, font=fnt, fill=255, align="center") # TO ALIGN CHARACTER IN CENTER
            #img = np.pad(img, pad_width=10, mode='constant', constant_values=0) #Manually added padding
            #img = Image.fromarray(img)
            img.save(path + "\\" + str(imageCounter) + "_" + str(repeats) + ".jpg")

Few sample results:少量样本结果：