人脸检测教程

---

目标

在本示例中，您将使用 Google Vision API 检测图片中的人脸。为了证明自己已正确检测到人脸，您将借助相应数据在每个人脸周围画一个方框。

查看代码示例：本教程中的大多数代码示例都摘自 GitHub 中更大的代码文件。点击代码示例上方的“在 GitHub 上查看”按钮，即可查看和下载该示例对应的完整源文件。

费用

在本文档中，您将使用 Google Cloud 的以下收费组件：

• Cloud Vision

您可使用价格计算器根据您的预计使用情况来估算费用。 Google Cloud 新用户可能有资格申请免费试用。

准备工作

1. Sign in to your Google Cloud account. If you're new to Google Cloud, create an account to evaluate how our products perform in real-world scenarios. New customers also get $300 in free credits to run, test, and deploy workloads.

2. In the Google Cloud console, on the project selector page, select or create a Google Cloud project.

   Note: If you don't plan to keep the resources that you create in this procedure, create a project instead of selecting an existing project. After you finish these steps, you can delete the project, removing all resources associated with the project.

   Go to project selector

3. Make sure that billing is enabled for your Google Cloud project.

4. Enable the Google Cloud Vision API.

   Enable the API

5. In the Google Cloud console, on the project selector page, select or create a Google Cloud project.

   Note: If you don't plan to keep the resources that you create in this procedure, create a project instead of selecting an existing project. After you finish these steps, you can delete the project, removing all resources associated with the project.

   Go to project selector

6. Make sure that billing is enabled for your Google Cloud project.

7. Enable the Google Cloud Vision API.

   Enable the API

8. 设置您的环境以便使用应用默认凭据。
9. 设置语言特定的任务和工具：

   C#

   • 安装 Google 客户端库。
   • 安装 Visual Studio 2012/2015。

   Java

   • 安装 Java。
   • API 参考文档。

   • 下载并安装 Apache Maven 构建系统。构建项目时，Maven 会确保安装 Google API 客户端库和 Vision API 客户端库，因为我们已将它们包含在 pom.xml 中。

     <dependency>
       <groupId>com.google.apis</groupId>
       <artifactId>google-api-services-vision</artifactId>
       <version>v1-rev20231219-2.0.0</version>
     </dependency>
     <dependency>
       <groupId>com.google.auth</groupId>
       <artifactId>google-auth-library-oauth2-http</artifactId>
     </dependency>
     <dependency>
       <groupId>com.google.guava</groupId>
       <artifactId>guava</artifactId>
     </dependency>
     <dependency>
       <groupId>com.google.http-client</groupId>
       <artifactId>google-http-client-jackson2</artifactId>
     </dependency>

     ...

   Node.js

   • 安装 Google 客户端库。
   • 安装 node.js。
   • API 参考。

   • 安装 npm 和 node-canvas。示例代码包含 package.json，用于使用 npm install 命令安装所有依赖项。请注意，您可能还需要安装 node-canvas 的其他依赖项；如需了解详情，请参阅 node-canvas 安装文档。

     {
       "name": "nodejs-docs-samples-vision",
       "private": true,
       "license": "Apache-2.0",
       "author": "Google LLC",
       "engines": {
         "node": ">=16.0.0"
       },
       "files": [
         "*.js"
       ],
       "scripts": {
         "test": "c8 mocha -p -j 2 system-test --timeout 600000"
       },
       "dependencies": {
         "@google-cloud/vision": "^4.0.0",
         "natural": "^8.0.0",
         "pureimage": "^0.3.17",
         "redis": "^4.6.5",
         "yargs": "^17.7.1"
       },
       "devDependencies": {
         "@google-cloud/storage": "^7.0.0",
         "@types/uuid": "^10.0.0",
         "@types/yargs": "^17.0.22",
         "c8": "^10.0.0",
         "chai": "^4.5.0",
         "mocha": "^10.2.0",
         "uuid": "^10.0.0"
       }
     }
     

   PHP

   • 安装 Google 客户端库。
   • 安装 PHP 和 Composer。
   • API 参考。
   • 安装并启用 PHP 版 GD。

   Python

   • 安装 Google 客户端库。
   • 安装 Python。
   • API 参考。
   • 安装 pillow。

   Ruby

   • 安装 Google 客户端库。
   • 安装 Ruby 和 Bundler。
   • API 参考。
   • 安装 ImageMagick。

创建服务对象

要使用官方客户端 SDK 访问 Google API，请根据相应 API 的发现文档（用于向 SDK 描述 API）来创建服务对象。您需要使用自己的凭据从 Vision API 的发现服务中获取该文档。

Java

import com.google.api.client.googleapis.javanet.GoogleNetHttpTransport;
import com.google.api.client.json.JsonFactory;
import com.google.api.client.json.gson.GsonFactory;
import com.google.api.services.vision.v1.Vision;
import com.google.api.services.vision.v1.VisionScopes;
import com.google.api.services.vision.v1.model.AnnotateImageRequest;
import com.google.api.services.vision.v1.model.AnnotateImageResponse;
import com.google.api.services.vision.v1.model.BatchAnnotateImagesRequest;
import com.google.api.services.vision.v1.model.BatchAnnotateImagesResponse;
import com.google.api.services.vision.v1.model.FaceAnnotation;
import com.google.api.services.vision.v1.model.Feature;
import com.google.api.services.vision.v1.model.Image;
import com.google.api.services.vision.v1.model.Vertex;
import com.google.auth.http.HttpCredentialsAdapter;
import com.google.auth.oauth2.GoogleCredentials;
import com.google.common.collect.ImmutableList;
import java.awt.BasicStroke;
import java.awt.Color;
import java.awt.Graphics2D;
import java.awt.Polygon;
import java.awt.image.BufferedImage;
import java.io.IOException;
import java.nio.file.Files;
import java.nio.file.Path;
import java.nio.file.Paths;
import java.security.GeneralSecurityException;
import java.util.List;
import javax.imageio.ImageIO;

/** Connects to the Vision API using Application Default Credentials. */
public static Vision getVisionService() throws IOException, GeneralSecurityException {
  GoogleCredentials credential =
      GoogleCredentials.getApplicationDefault().createScoped(VisionScopes.all());
  JsonFactory jsonFactory = GsonFactory.getDefaultInstance();
  return new Vision.Builder(
          GoogleNetHttpTransport.newTrustedTransport(),
          jsonFactory,
          new HttpCredentialsAdapter(credential))
      .setApplicationName(APPLICATION_NAME)
      .build();
}

Node.js

// By default, the client will authenticate using the service account file
// specified by the GOOGLE_APPLICATION_CREDENTIALS environment variable and use
// the project specified by the GCLOUD_PROJECT environment variable. See
// https://googlecloudplatform.github.io/gcloud-node/#/docs/google-cloud/latest/guides/authentication
const vision = require('@google-cloud/vision');

// Creates a client
const client = new vision.ImageAnnotatorClient();

const fs = require('fs');

Python

from google.cloud import vision
from PIL import Image, ImageDraw

client = vision.ImageAnnotatorClient()

发送人脸检测请求

要构建发送至 Vision API 的请求，请先查阅 API 文档。在本示例中，您将要求 images 资源对图片执行 annotate 操作。对此 API 的请求采用具有 requests 列表的对象形式。此列表中的每项均包含两种信息：

• Base64 编码的图片数据
• 要在该图片中注释的一列特征。

在本示例中，您只需针对一张图片请求 FACE_DETECTION 注释，然后返回相关的响应部分：

Java

/** Gets up to {@code maxResults} faces for an image stored at {@code path}. */
public List<FaceAnnotation> detectFaces(Path path, int maxResults) throws IOException {
  byte[] data = Files.readAllBytes(path);

  AnnotateImageRequest request =
      new AnnotateImageRequest()
          .setImage(new Image().encodeContent(data))
          .setFeatures(
              ImmutableList.of(
                  new Feature().setType("FACE_DETECTION").setMaxResults(maxResults)));
  Vision.Images.Annotate annotate =
      vision
          .images()
          .annotate(new BatchAnnotateImagesRequest().setRequests(ImmutableList.of(request)));
  // Due to a bug: requests to Vision API containing large images fail when GZipped.
  annotate.setDisableGZipContent(true);

  BatchAnnotateImagesResponse batchResponse = annotate.execute();
  assert batchResponse.getResponses().size() == 1;
  AnnotateImageResponse response = batchResponse.getResponses().get(0);
  if (response.getFaceAnnotations() == null) {
    throw new IOException(
        response.getError() != null
            ? response.getError().getMessage()
            : "Unknown error getting image annotations");
  }
  return response.getFaceAnnotations();
}

Node.js

async function detectFaces(inputFile) {
  // Make a call to the Vision API to detect the faces
  const request = {image: {source: {filename: inputFile}}};
  const results = await client.faceDetection(request);
  const faces = results[0].faceAnnotations;
  const numFaces = faces.length;
  console.log(`Found ${numFaces} face${numFaces === 1 ? '' : 's'}.`);
  return faces;
}

Python

def detect_face(face_file, max_results=4):
    """Uses the Vision API to detect faces in the given file.

    Args:
        face_file: A file-like object containing an image with faces.

    Returns:
        An array of Face objects with information about the picture.
    """
    client = vision.ImageAnnotatorClient()

    content = face_file.read()
    image = vision.Image(content=content)

    return client.face_detection(image=image, max_results=max_results).face_annotations

处理响应

恭喜！您已检测到图片中的人脸。对人脸注释请求的响应包含有关检测到的人脸的一些元数据，其中包括环绕人脸的多边形的坐标。不过目前，这只是一列数字。接下来，我们用这些数字来确认您确实已找到图片中的人脸。我们将使用 Vision API 返回的坐标在图片的副本上绘制多边形。

Java

/** Reads image {@code inputPath} and writes {@code outputPath} with {@code faces} outlined. */
private static void writeWithFaces(Path inputPath, Path outputPath, List<FaceAnnotation> faces)
    throws IOException {
  BufferedImage img = ImageIO.read(inputPath.toFile());
  annotateWithFaces(img, faces);
  ImageIO.write(img, "jpg", outputPath.toFile());
}

/** Annotates an image {@code img} with a polygon around each face in {@code faces}. */
public static void annotateWithFaces(BufferedImage img, List<FaceAnnotation> faces) {
  for (FaceAnnotation face : faces) {
    annotateWithFace(img, face);
  }
}

/** Annotates an image {@code img} with a polygon defined by {@code face}. */
private static void annotateWithFace(BufferedImage img, FaceAnnotation face) {
  Graphics2D gfx = img.createGraphics();
  Polygon poly = new Polygon();
  for (Vertex vertex : face.getFdBoundingPoly().getVertices()) {
    poly.addPoint(vertex.getX(), vertex.getY());
  }
  gfx.setStroke(new BasicStroke(5));
  gfx.setColor(new Color(0x00ff00));
  gfx.draw(poly);
}

Node.js

我们使用 node-canvas 库在图片上绘制多边形。

async function highlightFaces(inputFile, faces, outputFile, PImage) {
  // Open the original image
  const stream = fs.createReadStream(inputFile);
  let promise;
  if (inputFile.match(/\.jpg$/)) {
    promise = PImage.decodeJPEGFromStream(stream);
  } else if (inputFile.match(/\.png$/)) {
    promise = PImage.decodePNGFromStream(stream);
  } else {
    throw new Error(`Unknown filename extension ${inputFile}`);
  }
  const img = await promise;
  const context = img.getContext('2d');
  context.drawImage(img, 0, 0, img.width, img.height, 0, 0);

  // Now draw boxes around all the faces
  context.strokeStyle = 'rgba(0,255,0,0.8)';
  context.lineWidth = '5';

  faces.forEach(face => {
    context.beginPath();
    let origX = 0;
    let origY = 0;
    face.boundingPoly.vertices.forEach((bounds, i) => {
      if (i === 0) {
        origX = bounds.x;
        origY = bounds.y;
        context.moveTo(bounds.x, bounds.y);
      } else {
        context.lineTo(bounds.x, bounds.y);
      }
    });
    context.lineTo(origX, origY);
    context.stroke();
  });

  // Write the result to a file
  console.log(`Writing to file ${outputFile}`);
  const writeStream = fs.createWriteStream(outputFile);
  await PImage.encodePNGToStream(img, writeStream);
}

Python

def highlight_faces(image, faces, output_filename):
    """Draws a polygon around the faces, then saves to output_filename.

    Args:
      image: a file containing the image with the faces.
      faces: a list of faces found in the file. This should be in the format
          returned by the Vision API.
      output_filename: the name of the image file to be created, where the
          faces have polygons drawn around them.
    """
    im = Image.open(image)
    draw = ImageDraw.Draw(im)
    # Sepecify the font-family and the font-size
    for face in faces:
        box = [(vertex.x, vertex.y) for vertex in face.bounding_poly.vertices]
        draw.line(box + [box[0]], width=5, fill="#00ff00")
        # Place the confidence value/score of the detected faces above the
        # detection box in the output image
        draw.text(
            (
                (face.bounding_poly.vertices)[0].x,
                (face.bounding_poly.vertices)[0].y - 30,
            ),
            str(format(face.detection_confidence, ".3f")) + "%",
            fill="#FF0000",
        )
    im.save(output_filename)

综合应用

Java

/** Annotates an image using the Vision API. */
public static void main(String[] args) throws IOException, GeneralSecurityException {
  if (args.length != 2) {
    System.err.println("Usage:");
    System.err.printf(
        "\tjava %s inputImagePath outputImagePath\n", FaceDetectApp.class.getCanonicalName());
    System.exit(1);
  }
  Path inputPath = Paths.get(args[0]);
  Path outputPath = Paths.get(args[1]);
  if (!outputPath.toString().toLowerCase().endsWith(".jpg")) {
    System.err.println("outputImagePath must have the file extension 'jpg'.");
    System.exit(1);
  }

  FaceDetectApp app = new FaceDetectApp(getVisionService());
  List<FaceAnnotation> faces = app.detectFaces(inputPath, MAX_RESULTS);
  System.out.printf("Found %d face%s\n", faces.size(), faces.size() == 1 ? "" : "s");
  System.out.printf("Writing to file %s\n", outputPath);
  app.writeWithFaces(inputPath, outputPath, faces);
}

...

要构建并运行该示例，请从示例代码目录运行以下命令：

mvn clean compile assembly:single
java -cp target/vision-face-detection-1.0-SNAPSHOT-jar-with-dependencies.jar \
    com.google.cloud.vision.samples.facedetect.FaceDetectApp \
    data/face.jpg \
    output.jpg

Node.js

async function main(inputFile, outputFile) {
  const PImage = require('pureimage');
  outputFile = outputFile || 'out.png';
  const faces = await detectFaces(inputFile);
  console.log('Highlighting...');
  await highlightFaces(inputFile, faces, outputFile, PImage);
  console.log('Finished!');
}

要运行该示例，请从示例代码目录运行以下命令：

node faceDetection resources/face.png

Python

def main(input_filename, output_filename, max_results):
    with open(input_filename, "rb") as image:
        faces = detect_face(image, max_results)
        print("Found {} face{}".format(len(faces), "" if len(faces) == 1 else "s"))

        print(f"Writing to file {output_filename}")
        # Reset the file pointer, so we can read the file again
        image.seek(0)
        highlight_faces(image, faces, output_filename)

清理

为避免因本教程中使用的资源导致您的 Google Cloud 账号产生费用，请删除包含这些资源的项目，或者保留项目但删除各个资源。

注意：删除项目会产生以下影响：

• 项目中的所有内容都会被删除。如果您将现有项目用于本文档中的任务，则删除该项目后，还将删除您已在该项目中完成的任何其他工作。
• 自定义项目 ID 丢失。创建此项目时，您可能创建了要在将来使用的自定义项目 ID。要保留使用该项目 ID 的网址（例如 appspot.com 网址），请删除项目内的所选资源，而不是删除整个项目。

如果您打算探索多个架构、教程或快速入门，则重复使用项目可以帮助您避免超出项目配额限制。

1. In the Google Cloud console, go to the Manage resources page.

   Go to Manage resources

2. In the project list, select the project that you want to delete, and then click Delete.
3. In the dialog, type the project ID, and then click Shut down to delete the project.