前言

由于公司是2G，所以一些收费的公网api不能用（同时也不安全），以至于内部尝试了多种开源ocr框架。首先是使用golang封装的一个ocr模块gosseract，使用英文模型多数字字母识别准确率高一点，不过也只有80%多的准确率。后面就尝试用gunicorn+flask+PaddleOCR 简单开发了一个web服务。

gosseract（自己弄一个unbuntu的基础镜像）

dockerfile

RUN echo 'deb http://mirrors.163.com/ubuntu/ bionic main restricted universe multiverse \n\deb http://mirrors.163.com/ubuntu/ bionic-security main restricted universe multiverse \n\deb http://mirrors.163.com/ubuntu/ bionic-updates main restricted universe multiverse \n\deb http://mirrors.163.com/ubuntu/ bionic-proposed main restricted universe multiverse \n\deb http://mirrors.163.com/ubuntu/ bionic-backports main restricted universe multiverse \n\deb-src http://mirrors.163.com/ubuntu/ bionic main restricted universe multiverse \n\deb-src http://mirrors.163.com/ubuntu/ bionic-security main restricted universe multiverse \n\deb-src http://mirrors.163.com/ubuntu/ bionic-updates main restricted universe multiverse \n\deb-src http://mirrors.163.com/ubuntu/ bionic-proposed main restricted universe multiverse \n\deb-src http://mirrors.163.com/ubuntu/ bionic-backports main restricted universe multiverse' > /etc/apt/sources.listENV DEBIAN_FRONTEND noninteractive
RUN apt-get update && \apt-get -y install vim wget net-tools curl sudo make telnet iputils-ping tzdata git gcc libtbb2 zip && \ln -sf /usr/share/zoneinfo/Asia/Shanghai  /etc/localtime && echo "Asia/Shanghai" > /etc/timezoneRUN apt-get -y install automake ca-certificates g++ git libtool libleptonica-dev make pkg-config
RUN git clone https://github.com/tesseract-ocr/tesseract.git && cd tesseract && ./autogen.sh && ./configure && make && make install && ldconfig
#libleptonica 需要创建软连接才能使用
RUN ln -s  /usr/lib/x86_64-linux-gnu/liblept.so /usr/lib/x86_64-linux-gnu/libleptonica.so

 然后自己根基上述打一个基础镜像，自己的golang代码基于这个基础镜像来生成生产镜像。

gunicorn+flask+PaddleOCR

gunicorn是一个wcgi服务，类似网关和反向代理服务（参考php）。能够使用多进程的方式管理应用服务。

dockerfile（基础镜像）

FROM registry.baidubce.com/paddlepaddle/paddle:2.1.3-gpu-cuda10.2-cudnn7RUN echo 'deb http://mirrors.163.com/ubuntu/ bionic main restricted universe multiverse \n\deb http://mirrors.163.com/ubuntu/ bionic-security main restricted universe multiverse \n\deb http://mirrors.163.com/ubuntu/ bionic-updates main restricted universe multiverse \n\deb http://mirrors.163.com/ubuntu/ bionic-proposed main restricted universe multiverse \n\deb http://mirrors.163.com/ubuntu/ bionic-backports main restricted universe multiverse \n\deb-src http://mirrors.163.com/ubuntu/ bionic main restricted universe multiverse \n\deb-src http://mirrors.163.com/ubuntu/ bionic-security main restricted universe multiverse \n\deb-src http://mirrors.163.com/ubuntu/ bionic-updates main restricted universe multiverse \n\deb-src http://mirrors.163.com/ubuntu/ bionic-proposed main restricted universe multiverse \n\deb-src http://mirrors.163.com/ubuntu/ bionic-backports main restricted universe multiverse' > /etc/apt/sources.list
#不然的话会加载其他源 会报错
RUN rm -rf /etc/apt/apt.conf.d/* /etc/apt/sources.list.d/
# 不知道为什么原先的ssl居然不能用 太垃圾了（版本匹配不上）
RUN apt update && apt remove -y libssl-dev && apt install -y libssl-devRUN python3 -m pip install paddlepaddle "paddleocr>=2.0.1" -i https://mirror.baidu.com/pypi/simple
RUN pip3 install gunicorn gevent flask  -i https://mirror.baidu.com/pypi/simpleRUN echo "from paddleocr import PaddleOCR" >> download.py
# 预先加载英文模型 防止在代码跑起来之后加载 如果想要加载中文模型 就是复制俩行
RUN echo "PaddleOCR(use_angle_cls=True, lang=\"en\")" >> download.py
RUN python3 download.py

生产镜像

FROM ****/ocr_base:0.0.2WORKDIR /workspaceCOPY ./app/ocr/app.py /workspace/app.py
# 这里面是启动3个worker 不要太多 模型加载之后可是能消耗近2g物理内存
CMD cd /workspace && gunicorn -b 0.0.0.0:8000 -w 3 -k gevent --access-logfile - app:app

其中app.py 就是flask的入口文件

import time
import urllib.requestfrom flask import Flask, request
from paddleocr import PaddleOCR, draw_ocrdef save_image(url,outputfile):try:response = urllib.request.urlopen(url)data = response.read()with open(outputfile, "wb") as file:file.write(data)return Trueexcept urllib.error.URLError as e:print("Error occurred while retrieving the URL:", e)return Falseapp = Flask(__name__)
ocr = PaddleOCR(use_angle_cls=True, lang="en")@app.route("/")
def hello():print("-------------------")return "Hello World!"@app.post("/ocr/check")
def check_post():ret = {}req = request.get_json()print(req,type(req))url = req.get("url")if url == None :ret["code"] = -1ret["msg"] = "param url lost"ret["data"] = []return retresult = ocr.ocr(url, cls=True)if len(result) == 0:ret["code"] = -3ret["msg"] = "ocr result empty"ret["data"] = []return retdata =[]for idx in range(len(result)):res = result[idx]for idx1 in range(len(res)) :temp = {}res1 = res[idx1]temp["text"]=res1[-1][0]temp["score"]=res1[-1][1]data.append(temp)ret["code"] = 0ret["msg"] = ""ret["data"] = data    return retif __name__ == "__main__" :app.run()

到此你就搭建了一个ocr的web服务了

普通的验证码之类的识别1s 10张 想要更高的性能那你就在生产镜像里面吧worker加到更大，不过消耗的cpu内存也就更多（PaddleOCR其实支持gpu 这里默认是cpu）