好吧,当我用tesseract运行你的代码时,你提供的图像完美地返回了文本(包括破折号和空格)。在这一点上,你显然可以使用txt = txt.replace("-", "").replace(" ", "")
来摆脱破折号和空格。
Buuuuuut我知道OCR(即使与我们都使用tesseract)跨平台会有所不同,所以我已经包含了我的评论建议的示例。
首先,我们在划线分割的影像,然后我们读到每个分割图像,然后我们串接:
# I changed your imports a bit
from PIL import Image
from pyocr import pyocr
from pyocr import builders
import cStringIO
import os
# set up all your OCR stuff
os.putenv("TESSDATA_PREFIX", "/usr/share/tesseract-ocr/")
tools = pyocr.get_available_tools()
tool = tools[0]
langs = tool.get_available_languages()
lang = "eng" #set language to english to simplify things
# definte a function to return the text of a given image
def doOCR(fName):
txt = tool.image_to_string(Image.open(fName), lang=lang, builder=builders.TextBuilder())
return txt
# define the path of the image we are going to read
path = "test.png"
# get the image dimensions
im = Image.open(path)
width, height = im.size
# define the points we want to split the image at
# these are the points where the dashes are
split_points = [119, 158]
# define the file names for the image parts
split_names = ["split-1.png", "split-2.png", "split-3.png"]
# define a function to crop the image and remove the dashes
def doCrop(imagePath, cropPath, x, y, x2, y2):
im = Image.open(imagePath)
box = (x, y, x2, y2)
region = im.crop(box) # extract the box region
region.save(cropPath) # save it as a separate image
# in the image you provided each "-" is ~10 pixels long
lenpix = 10
# crop the image at the split points
doCrop(path, split_names[0], 0, 0, split_points[0], height) # get the first section
doCrop(path, split_names[1], split_points[0] + lenpix, 0, split_points[1], height) # get the middle section
doCrop(path, split_names[2], split_points[1] + lenpix, 0, width, height) # get the final section
# define a variable for our final value
finalValue = ""
# finally iterate through split files
# and add the OCR results from each split together
for f in split_names:
finalValue += doOCR(f) # concatenate the ocr value with the final
os.remove(f) # remove the split file now that we've used it
# display the final value
print finalValue
工作就像一个魅力对我来说:
希望这有助于!
“ - ”是否总是在同一个地方?如果是这种情况,您可以分割图像几次以删除“ - ”,然后连接每个单独图像上运行OCR的结果。 –
你能提供一个代码示例吗? – Sekai
已提供一些代码 - 请查看 –