PaddleOCR学习笔记2-初步识别服务

今天初步实现了网页，上传图片，识别显示结果到页面的服务。后续再完善。

采用flask + paddleocr+ bootstrap快速搭建OCR识别服务。

代码结构如下：

模板页面代码文件如下：

upload.html :

<!DOCTYPE html>
<html>
<meta charset="utf-8">
<head>
    <title>PandaCodeOCR</title>
    <!--静态加载 样式-->
    <link rel="stylesheet" href={{ url_for('static',filename='bootstrap3/css/bootstrap.min.css') }}></link>
    <style>
        body {
            font-family: Arial, sans-serif;
            margin: 0;
            padding: 0;
        }
        .header {
            background-color: #f0f0f0;
            text-align: center;
            padding: 20px;
        }
        .title {
            font-size: 32px;
            margin-bottom: 10px;
        }
 
        .menu {
            list-style-type: none;
            margin: 0;
            padding: 0;
            overflow: hidden;
            background-color: #FFDEAD;
            border: 2px solid #DCDCDC;
        }
 
        .menu li {
            float: left;
            font-size: 24px;
        }
 
        .menu li a {
            display: block;
            color: #333;
            text-align: center;
            padding: 14px 16px;
            text-decoration: none;
        }
 
        .menu li a:hover {
            background-color: #ddd;
        }
 
        .content {
            padding: 20px;
            border: 2px solid blue;
        }
    </style>
</head>
<body>
    <div class="header">
        <div class="title">PandaCodeOCR</div>
    </div>
 
    <ul class="menu">
        <li><a href="http://localhost:5000/uploader">通用文本识别</a></li>
    </ul>
 
    <div class="content">
        <!--上传图片文件-->
        <div id="upload_file">
            <form action="http://localhost:5000/uploader" method="POST" enctype="multipart/form-data">
                <div class="form-group">
                    <input type="file" class="form-control" id="upload_file" name="upload_file" placeholder="upload_file">
                </div>
                <div class="form-group">
                    <button type="submit" class="form-control btn-primary">上传图片文件</button>
                </div>
            </form>
        </div>
    </div>
</body>
</html>

result.html :

<!DOCTYPE html>
<html>
<meta charset="utf-8">
<head>
    <title>结果</title>
    <!--静态加载 样式-->
    <link rel="stylesheet" href={{ url_for('static',filename='bootstrap3/css/bootstrap.min.css') }}></link>
    <style>
        body {
            font-family: Arial, sans-serif;
            margin: 0;
            padding: 0;
        }
        .header {
            background-color: #f0f0f0;
            text-align: center;
            padding: 20px;
        }
        .title {
            font-size: 32px;
            margin-bottom: 10px;
        }
 
        .menu {
            list-style-type: none;
            margin: 0;
            padding: 0;
            overflow: hidden;
            background-color: #FFDEAD;
            border: 2px solid #DCDCDC;
        }
 
        .menu li {
            float: left;
            font-size: 24px;
        }
 
        .menu li a {
            display: block;
            color: #333;
            text-align: center;
            padding: 14px 16px;
            text-decoration: none;
        }
 
        .menu li a:hover {
            background-color: #ddd;
        }
    </style>
</head>
<body>
    <div class="header">
        <div class="title">PandaCodeOCR</div>
    </div>
 
    <ul class="menu">
        <li><a href="http://localhost:5000/uploader">通用文本识别</a></li>
    </ul>
 
    <div class="row">
            <!--显示上传的图片-->
            <div class="col-md-6" style="border: 2px solid #ddd;">
                <span class="label label-info">上传图片</span>
                <!--静态加载 图片-->
                <img src="{{ url_for('static', filename = result_dict['filename'])}}" alt="show_img"  class="img-responsive">
            </div>
 
            <div class="col-md-6" style="border: 2px solid #ddd;">
                <!--显示识别结果JSON报文列表-->
                <span class="label label-info">识别结果：</span>
                {% for line_str in result_dict['result'] %}
                    <p class="text-left">{{ line_str['text'] }}</p>
                {% endfor %}
            </div>
    </div>
</body>
</html>
<!--静态加载 script-->
<script src={{ url_for('static',filename='jquery1.3.3/jquery.min.js')}}></script>

主要视图代码文件如下：

views.py ：

100

101

102

103

104

105

106

107

108

109

110

111

import json
import os
import time
 
from . import blue_task
from flask import Flask, render_template, request
 
from paddleocr import PaddleOCR
from PIL import Image,ImageDraw
import numpy as np
 
'''
自定义模型测试ocr方法
'''
 
 
def test_model_ocr(img):
    # 返回字典结果对象
    result_dict = {'result': []}
    # paddleocr 目前支持的多语言语种可以通过修改lang参数进行切换
    # 例如`ch`, `en`, `fr`, `german`, `korean`, `japan`
    # 使用CPU预加载，不用GPU
    # 模型路径下必须包含model和params文件，目前开源的v3版本模型 已经是识别率很高的了
    # 还要更好的就要自己训练模型了。
    ocr = PaddleOCR(det_model_dir='./inference/ch_PP-OCRv3_det_infer/',
                    rec_model_dir='./inference/ch_PP-OCRv3_rec_infer/',
                    cls_model_dir='./inference/ch_ppocr_mobile_v2.0_cls_infer/',
                    use_angle_cls=True, lang="ch", use_gpu=False)
    # 识别图片文件
    result0 = ocr.ocr(img, cls=True)
    result = result0[0]
    for index in range(len(result)):
        line = result[index]
 
        tmp_dict = {}
        points = line[0]
        text = line[1][0]
        score = line[1][1]
        tmp_dict['points'] = points
        tmp_dict['text'] = text
        tmp_dict['score'] = score
 
        result_dict['result'].append(tmp_dict)
    return result_dict
 
# 转换图片
def convert_image(image, threshold=None):
    # 阈值 控制二值化程度，不能超过256，[200, 256]
    # 适当调大阈值，可以提高文本识别率，经过测试有效。
    if threshold is None:
        threshold = 200
    print('threshold : ', threshold)
    # 首先进行图片灰度处理
    image = image.convert("L")
    pixels = image.load()
    # 在进行二值化
    for x in range(image.width):
        for y in range(image.height):
            if pixels[x, y] > threshold:
                pixels[x, y] = 255
            else:
                pixels[x, y] = 0
    return image
 
@blue_task.route('/upload')
def upload_file():
    return render_template('upload.html')
 
@blue_task.route('/uploader', methods=['GET', 'POST'])
def uploader():
    if request.method == 'POST':
        #每个上传的文件首先会保存在服务器上的临时位置，然后将其实际保存到它的最终位置。
        filedata = request.files['upload_file']
        upload_filename = filedata.filename
        print(upload_filename)
        #保存文件到指定路径
        #目标文件的名称可以是硬编码的，也可以从 ​request.files[file] ​对象的​ filename ​属性中获取。
        #但是，建议使用 ​secure_filename()​ 函数获取它的安全版本
        img_path = os.path.join('upload/', upload_filename)
        filedata.save(img_path)
        print('file uploaded successfully')
 
        start = time.time()
 
        print('=======开始OCR识别======')
        # 打开图片
        img1 = Image.open(img_path)
        # 转换图片, 识别图片文本
        # print('转换图片，阈值=220时，再转换为ndarray数组, 识别图片文本')
        # 转换图片
        img2 = convert_image(img1, 220)
        # Image图像转换为ndarray数组
        img_2 = np.array(img2)
        # 识别图片
        result_dict = test_model_ocr(img_2)
 
        # 识别时间
        end = time.time()
        recognize_time = int((end - start) * 1000)
 
        result_dict["filename"] = img_path
        result_dict["recognize_time"] = str(recognize_time)
        result_dict["error_code"] = "000000"
        result_dict["error_msg"] = "识别成功"
 
        # return json.dumps(result_dict, ensure_ascii=False), {'Content-Type': 'application/json'}
        # render_template方法:渲染模板
        # 参数1: 模板名称  参数n: 传到模板里的数据
        return render_template('result.html', result_dict=result_dict)
    else:
        return render_template('upload.html')