.NET6实现基于PaddleOCR+OpenCV的OCRAPI

最近查阅资料，研究如何通过.net 6封装一个OCR API供自己的项目调用,期间了解到Google的Tesseract,尝试过后感觉Tesseract在中文识别的结果上达不到想要的预期,最后发现百度的开源项目PaddleOCR能满足需求,因此做如下尝试

新建解决方案OCRDemo,并创建Web Api项目OCR.API,类库项目OCR.Lib以及OCR.Application,项目结构如下:

图片

PaddleOCR和OpenCV在nuget都有封装好的包,直接安装如上图所示的nuget包,其中个别包的版本名称包含centos7-x64和win是为了分别应对不同平台的部署

安装好之后,在项目OCR.Application中创建接口IOCRService.cs

1
2
3
4
5
6
7
8
9
10
11
12
public interface IOCRService
{
    /// <summary>
    /// 识别
    /// </summary>
    /// <param name="picStream"></param>
    /// <param name="resizeLine"></param>
    /// <param name="cancellationToken"></param>
    /// <returns></returns>
    Task<OCRResult> RunAsync(Stream picStream, int resizeLine, CancellationToken cancellationToken);
}

以及IRotationDetectorService.cs

1
2
3
4
5
6
7
8
9
public interface IRotationService
{
    /// <summary>
    /// 旋转检测
    /// </summary>
    /// <param name="picStream"></param>
    /// <returns></returns>
    MemoryStream RotationCheck(Stream picStream);
}

创建OCRResult.cs,定义图形识别之后的返回结果

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
public class OCRResult
{
    public List<OCRRegion> Regions { get; set; }

    public string Text { get; set; }
}

public class OCRSize
{
    public double Height { get; set; }
    public double Width { get; set; }
}

public class  OCRPosition
{
    public double X { get; set; }
    public double Y { get; set; }
}

public class OCRRect
{
    public double Angle { get; set; }

    public OCRSize Size { get; set; }

    public OCRPosition Center { get; set; }
}

public class OCRRegion
{
    public double Score { get; set; }

    public OCRRect Rect { get; set; }

    public string Text { get; set; }
}

OCR.Lib项目中创建类Extensions.OpenCV.cs,实现OpenCV的部分图形预处理扩展方法

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
public static class Extensions
{
    /// <summary>
    /// 比例缩放大小
    /// </summary>
    /// <param name="src"></param>
    /// <param name="lineSize">宽+高的最大限定值</param>
    /// <returns></returns>
    public static Mat SrcResize(this Mat src, int lineSize)
    {
        //比例缩放
        int size = src.Width + src.Height;
        double wRate = src.Width / (double)size;
        double hRate = src.Height / (double)size;

        var reSize = new Size(wRate * lineSize, hRate * lineSize);

        using Mat resize_mat = new Mat();
        Cv2.Resize(src, resize_mat, reSize);
        src = resize_mat.Clone();
        return src;
    }

    //边缘检测

    //膨胀

    //腐蚀

    //二值化

    // ...
}

接着创建RotationService.cs实现IRotationService旋转检测并矫正旋转角度

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
public class RotationService : IRotationService
{
    public MemoryStream RotationCheck(Stream picStream)
    {
        using Mat src = Mat.FromStream(picStream, ImreadModes.Color);
        //旋转检测
        using PaddleRotationDetector detector = new PaddleRotationDetector(RotationDetectionModel.EmbeddedDefault);
        RotationResult rrotationResult = detector.Run(src);
        var rotation = rrotationResult.Rotation;

        if (rotation != RotationDegree._0)
        {
            //矫正旋转
            using Mat retotionSrc = rrotationResult.RestoreRotationInPlace(src).Clone();
            return retotionSrc.ToMemoryStream();
        }
        return src.ToMemoryStream();
    }
}

创建OCRService.cs实现IOCRService完成图形字符识别

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
public class OCRService : IOCRService
{
    private readonly QueuedPaddleOcrAll _paddleOcr;
    public OCRService(QueuedPaddleOcrAll paddleOcr)
    {
        _paddleOcr = paddleOcr;
    }

    /// <summary>
    /// 识别
    /// </summary>
    /// <param name="picStream"></param>
    /// <param name="resizeLine"></param>
    /// <param name="cancellationToken"></param>
    /// <returns></returns>
    public async Task<OCRResult> RunAsync(Stream picStream, int resizeLine, CancellationToken cancellationToken)
    {
        //获取原图
        using Mat src = Mat.FromStream(picStream, ImreadModes.Color);
        //比例缩放大小,以便识别
        using Mat resize_mat = src.SrcResize(resizeLine);
  
        var paddleResult = await _paddleOcr.Run(resize_mat, cancellationToken: cancellationToken);

        List<OCRRegion> regions = new ();
        foreach (var item in paddleResult.Regions)
        {
            regions.Add(new OCRRegion
            {
                Rect = new OCRRect
                {
                    Angle = item.Rect.Angle,
                    Center = new OCRPosition
                    {
                        X = item.Rect.Center.X,
                        Y = item.Rect.Center.Y
                    },
                    Size = new OCRSize
                    {
                        Height = item.Rect.Size.Height,
                        Width = item.Rect.Size.Width
                    }
                },
                Score = item.Score,
                Text = item.Text,
            });
        }

        return new OCRResult { 
             Text = paddleResult.Text,
             Regions = regions
        };
    }
}

创建Startup.cs通过DI注入服务

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
public static class Startup
{
    public static IServiceCollection AddOcrLib(this IServiceCollection services, IConfiguration config)
    {
        return services
            .AddTransient<IOCRService, OCRService>()
            .AddTransient<IRotationService, RotationService>()
            .AddPaddleOCR(config);
    }


    public static IServiceCollection AddPaddleOCR(this IServiceCollection services, IConfiguration config)
    {
        return services.AddSingleton(s =>
        {
            //GPU or CPU
            Action<PaddleConfig> device = config.GetSection("PaddleDevice").Value == "GPU" ? PaddleDevice.Gpu() : PaddleDevice.Mkldnn();
            return new QueuedPaddleOcrAll(() => new PaddleOcrAll(LocalFullModels.ChineseV3, device)
            {
                Enable180Classification = true,
                AllowRotateDetection = true,
            }, consumerCount: 1);
        });
    }
}

OCR.API项目中修改Program.cs,添加以下行:

1
builder.Services.AddOcrLib(builder.Configuration);

修改后如下:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
 
try
{
	var builder = WebApplication.CreateBuilder(args);

	// Add services to the container.

	builder.Services.AddControllers();

	builder.Services.AddOcrLib(builder.Configuration);

	var app = builder.Build();

	// Configure the HTTP request pipeline.

	app.UseStaticFiles();

	app.UseRouting();

	app.UseAuthorization();

	app.MapControllerRoute(
		name: "default",
		pattern: "Api/{controller=Home}/{action=Index}/{id?}");


	app.Run();

}
catch (Exception ex)
{
	throw ex;
}

appsettings.json文件中添加配置:

1
"PaddleDevice": "CPU" // or "GPU"

向Controllers文件夹下添加控制器OcrController.cs文件

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
[Route("Api/[controller]")]
[ApiController]
public class BaseApiController: ControllerBase
{
}

public class OcrController : BaseApiController
{
    private readonly IOCRService _ocrService;
    private readonly IRotationService _rotationSerivice;
    public OcrController(IOCRService ocrService,IRotationService rotaionService) 
    {
        _ocrService = ocrService;
        _rotationSerivice = rotaionService;
    }

    [HttpPost]
    [Route("run")]
    public async Task<OCRResult> RunAsync(IFormFile file,CancellationToken cancellationToken)
    {
        using Stream stream = file.OpenReadStream();

        try
        {
            //检测旋转
            var ms = _rotationSerivice.RotationCheck(stream);

            var result = await _ocrService.RunAsync(ms, 1500, cancellationToken);

            return result;
        }
        catch (Exception ex)
        {
            throw;
        }
    }
}

通过Postman上传一张身份证请求API查看返回结果如下:
图片

Dockerfile支持:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
#See https://aka.ms/containerfastmode to understand how Visual Studio uses this Dockerfile to build your images for faster debugging.

FROM sdflysha/dotnet6-focal-paddle2.2.2:latest AS base
WORKDIR /app
EXPOSE 80

FROM sdflysha/dotnet6-focal-paddle2.2.2:latest AS build
WORKDIR /src
COPY ["OCR.API/OCR.API.csproj", "OCR.API/"]
COPY ["OCR.Lib/OCR.Lib.csproj", "OCR.Lib/"]
COPY ["OCR.Application/OCR.Application.csproj", "OCR.Application/"]
RUN dotnet restore "OCR.API/OCR.API.csproj"
COPY . .
WORKDIR "/src/OCR.API"
RUN dotnet build "OCR.API.csproj" -c Release -o /app/build

FROM build AS publish
RUN dotnet publish "OCR.API.csproj" -c Release -o /app/publish /p:UseAppHost=false

FROM base AS final
WORKDIR /app
COPY --from=publish /app/publish .
ENTRYPOINT ["dotnet", "OCR.API.dll"]

最终解决方案的结构如下:
图片

最终的识别结果返回的文本内容格式需要结合实际需求进行相应的调整,并且为了提高文字识别的准确率可能还需要通过Opencv对原始图像进行一系列的预处理.

.NET 6
#PaddleOCR #OpenCV #.NET 6
.NET6实现基于PaddleOCR+OpenCV的OCRAPI
https://wangyuangen.github.io/2023/12/08/OCRDemoSolution/
作者
Yuangen Wang
发布于
2023年12月8日
许可协议