为什么无法使用PIL和pytesseract获取字符串？

Question

这是一个简单的Python 3光学字符识别（OCR）程序，用于获取字符串。我已经上传了目标gif文件，请下载并将其保存为/tmp/target.gif。

try:
    from PIL import Image
except ImportError:
    import Image
import pytesseract
print(pytesseract.image_to_string(Image.open('/tmp/target.gif')))

我把所有的错误信息都粘贴在这里，请修复它以获取图像中的字符。

/usr/lib/python3/dist-packages/PIL/Image.py:925: UserWarning: Couldn't allocate palette entry for transparency
  "for transparency")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/local/lib/python3.5/dist-packages/pytesseract/pytesseract.py", line 309, in image_to_string
    }[output_type]()
  File "/usr/local/lib/python3.5/dist-packages/pytesseract/pytesseract.py", line 308, in <lambda>
    Output.STRING: lambda: run_and_get_output(*args),
  File "/usr/local/lib/python3.5/dist-packages/pytesseract/pytesseract.py", line 208, in run_and_get_output
    temp_name, input_filename = save_image(image)
  File "/usr/local/lib/python3.5/dist-packages/pytesseract/pytesseract.py", line 136, in save_image
    image.save(input_file_name, format=img_extension, **image.info)
  File "/usr/lib/python3/dist-packages/PIL/Image.py", line 1728, in save
    save_handler(self, fp, filename)
  File "/usr/lib/python3/dist-packages/PIL/GifImagePlugin.py", line 407, in _save
    _get_local_header(fp, im, (0, 0), flags)
  File "/usr/lib/python3/dist-packages/PIL/GifImagePlugin.py", line 441, in _get_local_header
    transparency = int(transparency)
TypeError: int() argument must be a string, a bytes-like object or a number, not 'tuple'

我用bash中的convert命令将它转换。

convert  "/tmp/target.gif"   "/tmp/target.jpg"

我在这里展示/tmp/target.gif和/tmp/target.jpg。 然后再次执行上面的Python代码。

try:
    from PIL import Image
except ImportError:
    import Image
import pytesseract
print(pytesseract.image_to_string(Image.open('/tmp/target.jpg')))

使用pytesseract.image_to_string(Image.open('/tmp/target.jpg'))无法获取任何内容，只能得到空白字符。

对于Trenton_M的代码：

>>> img1 = remove_noise_and_smooth(r'/tmp/target.jpg')
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "<stdin>", line 3, in remove_noise_and_smooth
AttributeError: 'NoneType' object has no attribute 'astype'
Thalish Sajeed

对于Thalish Sajeed的代码：

省略由print(pytesseract.image_to_string(Image.open(filename)))导致的错误信息。

Type "help", "copyright", "credits" or "license" for more information.
>>> from PIL import Image
>>> import pytesseract
>>> import matplotlib.pyplot as plt
>>> import cv2
>>> import numpy as np
>>> 
>>> 
>>> def display_image(filename, length_box=60, width_box=30):
...     if type(filename) == np.ndarray:
...         image = filename
...     else:
...         image = cv2.imread(filename)
...     plt.figure(figsize=(length_box, width_box))
...     plt.imshow(image, cmap="gray")
... 
>>> 
>>> filename = r"/tmp/target.jpg"
>>> display_image(filename)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "<stdin>", line 7, in display_image
  File "/usr/local/lib/python3.5/dist-packages/matplotlib/pyplot.py", line 2699, in imshow
    None else {}), **kwargs)
  File "/usr/local/lib/python3.5/dist-packages/matplotlib/__init__.py", line 1810, in inner
    return func(ax, *args, **kwargs)
  File "/usr/local/lib/python3.5/dist-packages/matplotlib/axes/_axes.py", line 5494, in imshow
    im.set_data(X)
  File "/usr/local/lib/python3.5/dist-packages/matplotlib/image.py", line 634, in set_data
    raise TypeError("Image data cannot be converted to float")
TypeError: Image data cannot be converted to float
>>>

@Thalish Sajeed，为什么使用您的代码后我得到了9244K而不是0244k？ 这是我测试过的样本文件。

提取出的字符串。

@Trenton_M，请修正一下你的代码中存在的小错误和遗漏，并按照你的建议删除plt.show()这一行。

>>> import cv2,pytesseract
>>> import numpy as np
>>> import matplotlib.pyplot as plt
>>> 
>>> 
>>> def image_smoothening(img):
...     ret1, th1 = cv2.threshold(img, 88, 255, cv2.THRESH_BINARY)
...     ret2, th2 = cv2.threshold(th1, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
...     blur = cv2.GaussianBlur(th2, (5, 5), 0)
...     ret3, th3 = cv2.threshold(blur, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
...     return th3
... 
>>> 
>>> def remove_noise_and_smooth(file_name):
...     img = cv2.imread(file_name, 0)
...     filtered = cv2.adaptiveThreshold(img, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY, 9, 41)
...     kernel = np.ones((1, 1), np.uint8)
...     opening = cv2.morphologyEx(filtered, cv2.MORPH_OPEN, kernel)
...     closing = cv2.morphologyEx(opening, cv2.MORPH_CLOSE, kernel)
...     img = image_smoothening(img)
...     or_image = cv2.bitwise_or(img, closing)
...     return or_image
... 
>>> 
>>> cv2_thresh_list = [cv2.THRESH_BINARY, cv2.THRESH_TRUNC, cv2.THRESH_TOZERO]
>>> fn = r'/tmp/target.jpg'
>>> img1 = remove_noise_and_smooth(fn)
>>> img2 = cv2.imread(fn, 0)
>>> for i, img in enumerate([img1, img2]):
...     img_type = {0: 'Preprocessed Images\n',
...                 1: '\nUnprocessed Images\n'}
...     print(img_type[i])
...     for item in cv2_thresh_list:
...         print('Thresh: {}'.format(str(item)))
...         _, thresh = cv2.threshold(img, 127, 255, item)
...         plt.imshow(thresh, 'gray')
...         f_name = '{0}.jpg'.format(str(item))
...         plt.savefig(f_name)
...         print('OCR Result: {}\n'.format(pytesseract.image_to_string(f_name)))

预处理图像

在我的控制台中，所有输出信息如下：

Thresh: 0
<matplotlib.image.AxesImage object at 0x7fbc2519a6d8>
OCR Result: 10
15
20 

Edfifi
10
2 o 30 40 so
so

Thresh: 2
<matplotlib.image.AxesImage object at 0x7fbc255e7eb8>
OCR Result: 10
15
20
Edfifi
10
2 o 30 40 so
so
Thresh: 3
<matplotlib.image.AxesImage object at 0x7fbc25452fd0>
OCR Result: 10
15
20
Edfifi
10
2 o 30 40 so
so
Unprocessed Images
Thresh: 0
<matplotlib.image.AxesImage object at 0x7fbc25464c88>
OCR Result: 10
15
20
Thresh: 2
<matplotlib.image.AxesImage object at 0x7fbc254520f0>
OCR Result: 10
15
2o
2o
30 40 50
Thresh: 3
<matplotlib.image.AxesImage object at 0x7fbc1e1968d0>
OCR Result: 10
15
20

字符串0244R在哪里？

- showkey

Answer 1

让我们从JPG图像开始，因为pytesseract在GIF图像格式上操作时存在问题。参考文献

filename = "/tmp/target.jpg"
image = cv2.imread(filename)
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
ret, threshold = cv2.threshold(gray,55, 255, cv2.THRESH_BINARY)
print(pytesseract.image_to_string(threshold))

让我们尝试分解这些问题。

您的图像太嘈杂了，以至于tesseract引擎无法识别字母。我们使用一些简单的图像处理技术，如灰度和阈值处理，从图像中去除一些噪声。

然后，当我们将它发送到OCR引擎时，我们发现字母被更准确地捕获了。

如果您想要运行OCR，请查看商业OCR提供商，例如Google Cloud Vision。他们每月提供1000次免费OCR调用。您可以在此google-cloud-vision找到它们的链接。

您可以在我的笔记本中找到我测试这个的地方，如果您跟随此github link。

编辑 - 我已经使用了一些额外的图像清理技术更新了笔记本。源图像对于tesseract来说太嘈杂了，无法直接在图像上工作。您需要使用图像清理技术。您可以改变阈值参数或者使用其他技术替换高斯模糊，直到获得您想要的结果。

- Thalish Sajeed

---

让我们在聊天中继续这个讨论。 - Thalish Sajeed

Answer 2

首先，请确保您已安装Tesseract程序（不仅是Python包）。 解决方案的Jupyter笔记本：只有通过remove_noise_and_smooth传递的图像才能成功进行OCR转换。

尝试转换image.gif时，会生成 TypeError：int（）参数必须是字符串、类似于字节的对象或数字，而不是“tuple”。

将image.gif重命名为image.jpg，也会生成TypeError。

打开image.gif并将其"另存为"image.jpg，输出为空白，这意味着文本未被识别。

from PIL import Image
import pytesseract

# If you don't have tesseract executable in your PATH, include the following:
# your path may be different than mine
pytesseract.pytesseract.tesseract_cmd = "C:/Program Files (x86)/Tesseract-OCR/tesseract.exe"

imgo = Image.open('0244R_clean.jpg')

print(pytesseract.image_to_string(imgo))

• 无法从原始图像中识别出任何文本，因此可能需要在OCR之前进行后处理以清除。
• 我创建了一张干净的图片，pytesseract可以轻松地从中提取文本。下面包含了这张图片，你可以使用自己的代码测试它以验证功能。

添加后处理

使用图像预处理提高OCR准确性

OpenCV

import cv2
import numpy as np
import matplotlib.pyplot as plt


def image_smoothening(img):
    ret1, th1 = cv2.threshold(img, 88, 255, cv2.THRESH_BINARY)
    ret2, th2 = cv2.threshold(th1, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
    blur = cv2.GaussianBlur(th2, (5, 5), 0)
    ret3, th3 = cv2.threshold(blur, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
    return th3


def remove_noise_and_smooth(file_name):
    img = cv2.imread(file_name, 0)
    filtered = cv2.adaptiveThreshold(img, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY, 9, 41)
    kernel = np.ones((1, 1), np.uint8)
    opening = cv2.morphologyEx(filtered, cv2.MORPH_OPEN, kernel)
    closing = cv2.morphologyEx(opening, cv2.MORPH_CLOSE, kernel)
    img = image_smoothening(img)
    or_image = cv2.bitwise_or(img, closing)
    return or_image


cv2_thresh_list = [cv2.THRESH_BINARY, cv2.THRESH_TRUNC, cv2.THRESH_TOZERO]

fn = r'/tmp/target.jpg'
img1 = remove_noise_and_smooth(fn)
img2 = cv2.imread(fn, 0)
for i, img in enumerate([img1, img2]):
    img_type = {0: 'Preprocessed Images\n',
                1: '\nUnprocessed Images\n'}
    print(img_type[i])
    for item in cv2_thresh_list:
        print('Thresh: {}'.format(str(item)))
        _, thresh = cv2.threshold(img, 127, 255, item)
        plt.imshow(thresh, 'gray')
        f_name = '{}_{}.jpg'.format(i, str(item))
        plt.savefig(f_name)
        print('OCR Result: {}\n'.format(pytesseract.image_to_string(f_name)))

"img1会生成以下新图片："

img2将生成这些新图片：

- Trenton McKinney