利用Paddle OCR进行文字识别

利用Paddle OCR HubServing + Java Demo进行文字识别

需求描述

需要批量过滤一批本地图片，将含有网址水印的图片剔除

实现逻辑

使用python开源库 chinese_ocr_db_crnn_server 部署成本地http服务，使用Java进行http调用,获取图片中的文字信息，进行文字比较。

经测试，识别效果很好，准确率也很高，至于性能需要看服务器的配置了。

运行环境

名称	版本
操作系统	Centos7.9
python	3.7.0
pip3	22.1.1
paddlepaddle	2.3

1. 安装paddlepaddle

python -m pip install paddlepaddle==2.3.0 -i https://mirror.baidu.com/pypi/simple
## 验证安装是否成功,执行下面命令出现 PaddlePaddle is installed successfully 即是安装成功
python3
import paddle
paddle.utils.run_check()

安装的过程中由于硬件和操作系统环境不同，可能会出现某些依赖缺少的情况，需要自己进行排错安装

2. 下载Paddle OCR的源码

## git仓库克隆源码
git clone https://gitee.com/PaddlePaddle/PaddleOCR
## 移动源码到你想要的位置
mv ./PaddleOCR /PaddleOCR
##进入PaddleOCR目录进行依赖的安装
cd /PaddleOCR
pip3 install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple

3. 模型库下载

需要下载3个模型，分别是检测模型，方向分类器，识别模型

模型	链接
检测模型	https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_det_infer.tar
方向分类器	https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar
识别模型	https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_server_v2.0_rec_infer.tar

下载之后上传到PaddleOCR/interface目录下，interface如果无此目录则使用mkdir新建，然后解压

cd /PaddleOCR
mkdir interface
## 模型解压
tar -xvf ch_ppocr_mobile_v2.0_cls_infer.tar
tar -xvf ch_ppocr_server_v2.0_det_infer.tar
tar -xvf ch_ppocr_server_v2.0_rec_infer.tar

4. 安装HubServing服务

cd /PaddleOCR
export PYTHONPATH=.
## 下载服务模型
hub install chinese_ocr_db_crnn_server==1.1.1
## 服务启动
nohup hub serving start -m chinese_ocr_db_crnn_server >> /PaddleOCR/hub.log  2>&1 &
##出现 Running on http:/127.0.0.1:8866之类的信息表示启动成功
## 服务停止命令
kill -9 pid
## 或者
hub serving stop

5. 编写Java单元测试

package com.example.test.util;import java.io.*;
import java.util.ArrayList;
import java.util.List;/*** @autor Hou Dehong* @date 2022/5/19* @description*/
public class PaddleDemo2 {public static List<String> list = new ArrayList<>();public static void main(String[] args) throws Exception {String url = "http://10.10.102.208:8866/predict/chinese_ocr_db_crnn_server";String dirOri = "C:\\Users\\Administrator\\Desktop\\";String fileName = "test.jpg";// 读取文件byte[] imgData = readFileByBytes(dirOri + fileName);String imgStr = Base64Util.encode(imgData);String param = "{\"images\":[\"" + imgStr + "\"]}";System.out.println(fileName);String result = HttpUtil.postGeneralUrl(url, "application/json", param,"UTF-8");System.out.println(result);}public static byte[] readFileByBytes(String filePath) throws IOException {File file = new File(filePath);if (!file.exists()) {throw new FileNotFoundException(filePath);} else {ByteArrayOutputStream bos = new ByteArrayOutputStream((int) file.length());BufferedInputStream in = null;try {in = new BufferedInputStream(new FileInputStream(file));short bufSize = 1024;byte[] buffer = new byte[bufSize];int len1;while (-1 != (len1 = in.read(buffer, 0, bufSize))) {bos.write(buffer, 0, len1);}byte[] var7 = bos.toByteArray();return var7;} finally {try {if (in != null) {in.close();}} catch (IOException var14) {var14.printStackTrace();}bos.close();}}}
}

package com.example.test.util;/*** Base64 工具类*/
public class Base64Util {private static final char last2byte = (char) Integer.parseInt("00000011", 2);private static final char last4byte = (char) Integer.parseInt("00001111", 2);private static final char last6byte = (char) Integer.parseInt("00111111", 2);private static final char lead6byte = (char) Integer.parseInt("11111100", 2);private static final char lead4byte = (char) Integer.parseInt("11110000", 2);private static final char lead2byte = (char) Integer.parseInt("11000000", 2);private static final char[] encodeTable = new char[]{'A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J', 'K', 'L', 'M', 'N', 'O', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X', 'Y', 'Z', 'a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', 'x', 'y', 'z', '0', '1', '2', '3', '4', '5', '6', '7', '8', '9', '+', '/'};public Base64Util() {}public static String encode(byte[] from) {StringBuilder to = new StringBuilder((int) ((double) from.length * 1.34D) + 3);int num = 0;char currentByte = 0;int i;for (i = 0; i < from.length; ++i) {for (num %= 8; num < 8; num += 6) {switch (num) {case 0:currentByte = (char) (from[i] & lead6byte);currentByte = (char) (currentByte >>> 2);case 1:case 3:case 5:default:break;case 2:currentByte = (char) (from[i] & last6byte);break;case 4:currentByte = (char) (from[i] & last4byte);currentByte = (char) (currentByte << 2);if (i + 1 < from.length) {currentByte = (char) (currentByte | (from[i + 1] & lead2byte) >>> 6);}break;case 6:currentByte = (char) (from[i] & last2byte);currentByte = (char) (currentByte << 4);if (i + 1 < from.length) {currentByte = (char) (currentByte | (from[i + 1] & lead4byte) >>> 4);}}to.append(encodeTable[currentByte]);}}if (to.length() % 4 != 0) {for (i = 4 - to.length() % 4; i > 0; --i) {to.append("=");}}return to.toString();}
}

package com.example.test.util;import java.io.BufferedReader;
import java.io.DataOutputStream;
import java.io.InputStreamReader;
import java.net.HttpURLConnection;
import java.net.URL;
import java.util.List;
import java.util.Map;/*** http 工具类*/
public class HttpUtil {public static String post(String requestUrl, String accessToken, String params)throws Exception {//String contentType = "application/x-www-form-urlencoded";String contentType = "application/json";return HttpUtil.post(requestUrl, accessToken, contentType, params);}public static String post(String requestUrl, String accessToken, String contentType, String params)throws Exception {String encoding = "UTF-8";if (requestUrl.contains("nlp")) {encoding = "GBK";}return HttpUtil.post(requestUrl, accessToken, contentType, params, encoding);}public static String post(String requestUrl, String accessToken, String contentType, String params, String encoding)throws Exception {// String url = requestUrl + "?access_token=" + accessToken;String url = requestUrl;return HttpUtil.postGeneralUrl(url, contentType, params, encoding);}public static String postGeneralUrl(String generalUrl, String contentType, String params, String encoding)throws Exception {URL url = new URL(generalUrl);// 打开和URL之间的连接HttpURLConnection connection = (HttpURLConnection) url.openConnection();connection.setRequestMethod("POST");// 设置通用的请求属性connection.setRequestProperty("Content-Type", contentType);// connection.setRequestProperty("Connection", "Keep-Alive");connection.setUseCaches(false);connection.setDoOutput(true);connection.setDoInput(true);// 得到请求的输出流对象DataOutputStream out = new DataOutputStream(connection.getOutputStream());out.write(params.getBytes(encoding));out.flush();out.close();// 建立实际的连接connection.connect();// 获取所有响应头字段Map<String, List<String>> headers = connection.getHeaderFields();// 遍历所有的响应头字段/* for (String key : headers.keySet()) {System.err.println(key + "--->" + headers.get(key));}*/// 定义 BufferedReader输入流来读取URL的响应BufferedReader in = null;in = new BufferedReader(new InputStreamReader(connection.getInputStream(), encoding));String result = "";String getLine;while ((getLine = in.readLine()) != null) {result += getLine;}in.close();// System.err.println("result:" + result);return result;}
}

6.调用返回结果

text中的即为识别出的文字，文字识别出来了，可以在java中进行text的获取判断，实现自己想实现的需求

{"msg":"","results":[{"data":[{"confidence":0.9984913468360901,"text":"只有自己变优秀了","text_box_position":[[182,357],[391,357],[391,380],[182,380]]},{"confidence":0.9871423244476318,"text":"其他的事情才会跟着好起来","text_box_position":[[181,391],[501,391],[501,414],[181,414]]}],"save_path":""}],"status":"000"}

测试图片

[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-59HUgBiE-1656496465758)(C:\Users\Administrator\AppData\Roaming\Typora\typora-user-images\image-20220629174229948.png)]

9984913468360901,“text”:“只有自己变优秀了”,“text_box_position”:[[182,357],[391,357],[391,380],[182,380]]},{“confidence”:0.9871423244476318,“text”:“其他的事情才会跟着好起来”,“text_box_position”:[[181,391],[501,391],[501,414],[181,414]]}],“save_path”:“”}],“status”:“000”}


测试图片
![在这里插入图片描述](https://img-blog.csdnimg.cn/a07158541f3449169bf7acb7ba7ee592.jpeg#pic_center)