一、需求分析

产品经理召集开会，文章审核功能已经交付了，文章也能正常发布审核。对于上次提出的自管理敏感词也很满意，这次会议核心的内容如下：

文章中包含的图片要识别文字，过滤掉图片文字的敏感词

二、图片文字识别

什么是OCR?

OCR （Optical Character Recognition，光学字符识别）是指电子设备（例如扫描仪或数码相机）检查纸上打印的字符，通过检测暗、亮的模式确定其形状，然后用字符识别方法将形状翻译成计算机文字的过程

方案	说明
百度OCR	收费
Tesseract-OCR	Google维护的开源OCR引擎，支持Java，Python等语言调用
Tess4J	封装了Tesseract-OCR ，支持Java调用

三、Tess4j案例

①：创建项目导入tess4j对应的依赖

<dependency>
    <groupId>net.sourceforge.tess4j</groupId>
    <artifactId>tess4j</artifactId>
    <version>4.1.1</version>
</dependency>

②：导入中文字体库，把资料中的tessdata文件夹拷贝到自己的工作空间下

③：编写测试类进行测试

package com.heima.tess4j;

import net.sourceforge.tess4j.ITesseract;
import net.sourceforge.tess4j.Tesseract;

import java.io.File;

public class Application {

    public static void main(String[] args) {
        try {
            //获取本地图片
            File file = new File("D:\\26.png");
            //创建Tesseract对象
            ITesseract tesseract = new Tesseract();
            //设置字体库路径
            tesseract.setDatapath("D:\\workspace\\tessdata");
            //中文识别
            tesseract.setLanguage("chi_sim");
            //执行ocr识别
            String result = tesseract.doOCR(file);
            //替换回车和tal键  使结果为一行
            result = result.replaceAll("\\r|\\n","-").replaceAll(" ","");
            System.out.println("识别的结果为："+result);
        } catch (Exception e) {
            e.printStackTrace();
        }
    }
}

四、管理敏感词和图片文字识别集成到文章审核

①：在heima-leadnews-common中创建工具类，简单封装一下tess4j

需要先导入pom

<dependency>
    <groupId>net.sourceforge.tess4j</groupId>
    <artifactId>tess4j</artifactId>
    <version>4.1.1</version>
</dependency>

工具类

package com.heima.common.tess4j;

import lombok.Getter;
import lombok.Setter;
import net.sourceforge.tess4j.ITesseract;
import net.sourceforge.tess4j.Tesseract;
import net.sourceforge.tess4j.TesseractException;
import org.springframework.boot.context.properties.ConfigurationProperties;
import org.springframework.stereotype.Component;

import java.awt.image.BufferedImage;

@Getter
@Setter
@Component
@ConfigurationProperties(prefix = "tess4j")
public class Tess4jClient {

    private String dataPath;
    private String language;

    public String doOCR(BufferedImage image) throws TesseractException {
        //创建Tesseract对象
        ITesseract tesseract = new Tesseract();
        //设置字体库路径
        tesseract.setDatapath(dataPath);
        //中文识别
        tesseract.setLanguage(language);
        //执行ocr识别
        String result = tesseract.doOCR(image);
        //替换回车和tal键  使结果为一行
        result = result.replaceAll("\\r|\\n", "-").replaceAll(" ", "");
        return result;
    }

}

在spring.factories配置中添加该类,完整如下：

org.springframework.boot.autoconfigure.EnableAutoConfiguration=\
  com.heima.common.exception.ExceptionCatch,\
  com.heima.common.swagger.SwaggerConfiguration,\
  com.heima.common.swagger.Swagger2Configuration,\
  com.heima.common.aliyun.GreenTextScan,\
  com.heima.common.aliyun.GreenImageScan,\
  com.heima.common.tess4j.Tess4jClient

②：在heima-leadnews-wemedia中的配置中添加两个属性（字图库的位置和前缀名）

tess4j:
  data-path: D:\workspace\tessdata
  language: chi_sim

③：在WmNewsAutoScanServiceImpl中的handleImageScan方法上添加如下代码

try {
    for (String image : images) {
        byte[] bytes = fileStorageService.downLoadFile(image);

        //图片识别文字审核---begin-----

        //从byte[]转换为butteredImage
        ByteArrayInputStream in = new ByteArrayInputStream(bytes);
        BufferedImage imageFile = ImageIO.read(in);
        //识别图片的文字
        String result = tess4jClient.doOCR(imageFile);

        //审核是否包含自管理的敏感词
        boolean isSensitive = handleSensitiveScan(result, wmNews);
        if(!isSensitive){
            return isSensitive;
        }

        //图片识别文字审核---end-----


        imageList.add(bytes);

    } 
}catch (Exception e){
    e.printStackTrace();
}

最后附上文章审核的完整代码如下：

package com.heima.wemedia.service.impl;

import com.alibaba.fastjson.JSONArray;
import com.baomidou.mybatisplus.core.toolkit.Wrappers;
import com.heima.apis.article.IArticleClient;
import com.heima.common.aliyun.GreenImageScan;
import com.heima.common.aliyun.GreenTextScan;
import com.heima.common.tess4j.Tess4jClient;
import com.heima.file.service.FileStorageService;
import com.heima.model.article.dtos.ArticleDto;
import com.heima.model.common.dtos.ResponseResult;
import com.heima.model.wemedia.pojos.WmChannel;
import com.heima.model.wemedia.pojos.WmNews;
import com.heima.model.wemedia.pojos.WmSensitive;
import com.heima.model.wemedia.pojos.WmUser;
import com.heima.utils.common.SensitiveWordUtil;
import com.heima.wemedia.mapper.WmChannelMapper;
import com.heima.wemedia.mapper.WmNewsMapper;
import com.heima.wemedia.mapper.WmSensitiveMapper;
import com.heima.wemedia.mapper.WmUserMapper;
import com.heima.wemedia.service.WmNewsAutoScanService;
import lombok.extern.slf4j.Slf4j;
import org.apache.commons.lang3.StringUtils;
import org.springframework.beans.BeanUtils;
import org.springframework.beans.factory.annotation.Autowired;
import org.springframework.scheduling.annotation.Async;
import org.springframework.stereotype.Service;
import org.springframework.transaction.annotation.Transactional;

import javax.imageio.ImageIO;
import java.awt.image.BufferedImage;
import java.io.ByteArrayInputStream;
import java.util.*;
import java.util.stream.Collectors;


@Service
@Slf4j
@Transactional
public class WmNewsAutoScanServiceImpl implements WmNewsAutoScanService {

    @Autowired
    private WmNewsMapper wmNewsMapper;

    /**
     * 自媒体文章审核
     *
     * @param id 自媒体文章id
     */
    @Override
    @Async  //标明当前方法是一个异步方法
    public void autoScanWmNews(Integer id) {

//        int a = 1/0;

        //1.查询自媒体文章
        WmNews wmNews = wmNewsMapper.selectById(id);
        if (wmNews == null) {
            throw new RuntimeException("WmNewsAutoScanServiceImpl-文章不存在");
        }

        if (wmNews.getStatus().equals(WmNews.Status.SUBMIT.getCode())) {
            //从内容中提取纯文本内容和图片
            Map<String, Object> textAndImages = handleTextAndImages(wmNews);

            //自管理的敏感词过滤
            boolean isSensitive = handleSensitiveScan((String) textAndImages.get("content"), wmNews);
            if(!isSensitive) return;

            //2.审核文本内容  阿里云接口
            boolean isTextScan = handleTextScan((String) textAndImages.get("content"), wmNews);
            if (!isTextScan) return;

            //3.审核图片  阿里云接口
            boolean isImageScan = handleImageScan((List<String>) textAndImages.get("images"), wmNews);
            if (!isImageScan) return;

            //4.审核成功，保存app端的相关的文章数据
            ResponseResult responseResult = saveAppArticle(wmNews);
            if (!responseResult.getCode().equals(200)) {
                throw new RuntimeException("WmNewsAutoScanServiceImpl-文章审核，保存app端相关文章数据失败");
            }
            //回填article_id
            wmNews.setArticleId((Long) responseResult.getData());
            updateWmNews(wmNews, (short) 9, "审核成功");

        }
    }

    @Autowired
    private WmSensitiveMapper wmSensitiveMapper;

    /**
     * 自管理的敏感词审核
     * @param content
     * @param wmNews
     * @return
     */
    private boolean handleSensitiveScan(String content, WmNews wmNews) {

        boolean flag = true;

        //获取所有的敏感词
        List<WmSensitive> wmSensitives = wmSensitiveMapper.selectList(Wrappers.<WmSensitive>lambdaQuery().select(WmSensitive::getSensitives));
        List<String> sensitiveList = wmSensitives.stream().map(WmSensitive::getSensitives).collect(Collectors.toList());

        //初始化敏感词库
        SensitiveWordUtil.initMap(sensitiveList);

        //查看文章中是否包含敏感词
        Map<String, Integer> map = SensitiveWordUtil.matchWords(content);
        if(map.size() >0){
            updateWmNews(wmNews,(short) 2,"当前文章中存在违规内容"+map);
            flag = false;
        }

        return flag;
    }

    @Autowired
    private IArticleClient articleClient;

    @Autowired
    private WmChannelMapper wmChannelMapper;

    @Autowired
    private WmUserMapper wmUserMapper;

    /**
     * 保存app端相关的文章数据
     *
     * @param wmNews
     */
    private ResponseResult saveAppArticle(WmNews wmNews) {

        ArticleDto dto = new ArticleDto();
        //属性的拷贝
        BeanUtils.copyProperties(wmNews, dto);
        //文章的布局
        dto.setLayout(wmNews.getType());
        //频道
        WmChannel wmChannel = wmChannelMapper.selectById(wmNews.getChannelId());
        if (wmChannel != null) {
            dto.setChannelName(wmChannel.getName());
        }

        //作者
        dto.setAuthorId(wmNews.getUserId().longValue());
        WmUser wmUser = wmUserMapper.selectById(wmNews.getUserId());
        if (wmUser != null) {
            dto.setAuthorName(wmUser.getName());
        }

        //设置文章id
        if (wmNews.getArticleId() != null) {
            dto.setId(wmNews.getArticleId());
        }
        dto.setCreatedTime(new Date());

        ResponseResult responseResult = articleClient.saveArticle(dto);
        return responseResult;

    }


    @Autowired
    private FileStorageService fileStorageService;

    @Autowired
    private GreenImageScan greenImageScan;

    @Autowired
    private Tess4jClient tess4jClient;

    /**
     * 审核图片
     *
     * @param images
     * @param wmNews
     * @return
     */
    private boolean handleImageScan(List<String> images, WmNews wmNews) {

        boolean flag = true;

        if (images == null || images.size() == 0) {
            return flag;
        }

        //下载图片 minIO
        //图片去重
        images = images.stream().distinct().collect(Collectors.toList());

        List<byte[]> imageList = new ArrayList<>();

        try {
            for (String image : images) {
                byte[] bytes = fileStorageService.downLoadFile(image);

                //图片识别文字审核---begin-----

                //从byte[]转换为butteredImage
                ByteArrayInputStream in = new ByteArrayInputStream(bytes);
                BufferedImage imageFile = ImageIO.read(in);
                //识别图片的文字
                String result = tess4jClient.doOCR(imageFile);

                //审核是否包含自管理的敏感词
                boolean isSensitive = handleSensitiveScan(result, wmNews);
                if(!isSensitive){
                    return isSensitive;
                }

                //图片识别文字审核---end-----


                imageList.add(bytes);

            }
        }catch (Exception e){
            e.printStackTrace();
        }


        //审核图片
        try {
            Map map = greenImageScan.imageScan(imageList);
            if (map != null) {
                //审核失败
                if (map.get("suggestion").equals("block")) {
                    flag = false;
                    updateWmNews(wmNews, (short) 2, "当前文章中存在违规内容");
                }

                //不确定信息  需要人工审核
                if (map.get("suggestion").equals("review")) {
                    flag = false;
                    updateWmNews(wmNews, (short) 3, "当前文章中存在不确定内容");
                }
            }

        } catch (Exception e) {
            flag = false;
            e.printStackTrace();
        }
        return flag;
    }

    @Autowired
    private GreenTextScan greenTextScan;

    /**
     * 审核纯文本内容
     *
     * @param content
     * @param wmNews
     * @return
     */
    private boolean handleTextScan(String content, WmNews wmNews) {

        boolean flag = true;

        if ((wmNews.getTitle() + "-" + content).length() == 0) {
            return flag;
        }
        try {
            Map map = greenTextScan.greeTextScan((wmNews.getTitle() + "-" + content));
            if (map != null) {
                //审核失败
                if (map.get("suggestion").equals("block")) {
                    flag = false;
                    updateWmNews(wmNews, (short) 2, "当前文章中存在违规内容");
                }

                //不确定信息  需要人工审核
                if (map.get("suggestion").equals("review")) {
                    flag = false;
                    updateWmNews(wmNews, (short) 3, "当前文章中存在不确定内容");
                }
            }
        } catch (Exception e) {
            flag = false;
            e.printStackTrace();
        }

        return flag;

    }

    /**
     * 修改文章内容
     *
     * @param wmNews
     * @param status
     * @param reason
     */
    private void updateWmNews(WmNews wmNews, short status, String reason) {
        wmNews.setStatus(status);
        wmNews.setReason(reason);
        wmNewsMapper.updateById(wmNews);
    }

    /**
     * 1。从自媒体文章的内容中提取文本和图片
     * 2.提取文章的封面图片
     *
     * @param wmNews
     * @return
     */
    private Map<String, Object> handleTextAndImages(WmNews wmNews) {

        //存储纯文本内容
        StringBuilder stringBuilder = new StringBuilder();

        List<String> images = new ArrayList<>();

        //1。从自媒体文章的内容中提取文本和图片
        if (StringUtils.isNotBlank(wmNews.getContent())) {
            List<Map> maps = JSONArray.parseArray(wmNews.getContent(), Map.class);
            for (Map map : maps) {
                if (map.get("type").equals("text")) {
                    stringBuilder.append(map.get("value"));
                }

                if (map.get("type").equals("image")) {
                    images.add((String) map.get("value"));
                }
            }
        }
        //2.提取文章的封面图片
        if (StringUtils.isNotBlank(wmNews.getImages())) {
            String[] split = wmNews.getImages().split(",");
            images.addAll(Arrays.asList(split));
        }

        Map<String, Object> resultMap = new HashMap<>();
        resultMap.put("content", stringBuilder.toString());
        resultMap.put("images", images);
        return resultMap;

    }
}