使用 Pytesseract 进行文本定位、检测和识别

Pytesseract或Python-tesseract是Python的光学字符识别 (OCR) 工具。它将读取和识别图像、车牌等中的文本。Python-tesseract 实际上是Google 的 Tesseract-OCR Engine的包装类或包。它也很有用，被视为 tesseract 的独立调用脚本，因为它可以轻松读取Pillow和Leptonica成像库支持的所有图像类型，主要包括 -

• jpg
• PNG
• gif
• bmp
• tiff 等

此外，如果将其用作脚本，Python-tesseract 还将打印识别的文本，而不是将其写入文件。可以使用 pip 安装 Python-tesseract，如下所示 -

pip install pytesseract

如果您使用的是 Anaconda Cloud，可以安装 Python-tesseract，如下所示：-

conda install -c conda-forge/label/cf202003 pytesseract

或者

conda install -c conda-forge pytesseract

注意：在运行以下脚本之前，应该在系统中安装 tesseract。
下面是实现。

Python3

from pytesseract import*
import argparse
import cv2
 
 
# We construct the argument parser
# and parse the arguments
ap = argparse.ArgumentParser()
 
ap.add_argument("-i", "--image",
                required=True,
                help="path to input image to be OCR'd")
ap.add_argument("-c", "--min-conf",
                type=int, default=0,
                help="minimum confidence value to filter weak text detection")
args = vars(ap.parse_args())
 
# We load the input image and then convert
# it to RGB from BGR. We then use Tesseract
# to localize each area of text in the input
# image
images = cv2.imread(args["image"])
rgb = cv2.cvtColor(images, cv2.COLOR_BGR2RGB)
results = pytesseract.image_to_data(rgb, output_type=Output.DICT)
 
# Then loop over each of the individual text
# localizations
for i in range(0, len(results["text"])):
     
    # We can then extract the bounding box coordinates
    # of the text region from  the current result
    x = results["left"][i]
    y = results["top"][i]
    w = results["width"][i]
    h = results["height"][i]
     
    # We will also extract the OCR text itself along
    # with the confidence of the text localization
    text = results["text"][i]
    conf = int(results["conf"][i])
     
    # filter out weak confidence text localizations
    if conf > args["min_conf"]:
         
        # We will display the confidence and text to
        # our terminal
        print("Confidence: {}".format(conf))
        print("Text: {}".format(text))
        print("")
         
        # We then strip out non-ASCII text so we can
        # draw the text on the image We will be using
        # OpenCV, then draw a bounding box around the
        # text along with the text itself
        text = "".join(text).strip()
        cv2.rectangle(images,
                      (x, y),
                      (x + w, y + h),
                      (0, 0, 255), 2)
        cv2.putText(images,
                    text,
                    (x, y - 10),
                    cv2.FONT_HERSHEY_SIMPLEX,
                    1.2, (0, 255, 255), 3)
         
# After all, we will show the output image
cv2.imshow("Image", images)
cv2.waitKey(0)

输出：
执行以下命令查看输出

python ocr.py --image ocr.png 

除了输出，我们将看到置信度和命令提示符中的文本，如下所示 -

Confidence: 93
Text: I

Confidence: 93
Text: LOVE

Confidence: 91
Text: TESSERACT