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。

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. 登录您的 Google Cloud 账号。如果您是 Google Cloud 新手，请创建一个账号来评估我们的产品在实际场景中的表现。新客户还可获享 $300 赠金，用于运行、测试和部署工作负载。

  2. 在 Google Cloud Console 中的项目选择器页面上，选择或创建一个 Google Cloud 项目。

    转到“项目选择器”

  3. 确保您的 Google Cloud 项目已启用结算功能。

  4. 启用 Google Cloud Vision API。

    启用 API

    5.

  5. 在 Google Cloud Console 中的项目选择器页面上，选择或创建一个 Google Cloud 项目。

    转到“项目选择器”

  6. 确保您的 Google Cloud 项目已启用结算功能。

  7. 启用 Google Cloud Vision API。

    启用 API

    8.
  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#

    Java

    • 安装 Java。
    • API 参考。

    • 下载并安装 Apache Maven 构建系统。构建项目时，Maven 将确保安装 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": "^6.2.0",
    "pureimage": "^0.3.17",
    "redis": "^4.6.5",
    "yargs": "^17.7.1"
  },
  "devDependencies": {
    "@google-cloud/storage": "^7.0.0",
    "@types/uuid": "^9.0.1",
    "@types/yargs": "^17.0.22",
    "c8": "^8.0.0",
    "chai": "^4.3.7",
    "mocha": "^10.2.0",
    "uuid": "^9.0.0"
  }
}

    PHP

    Python

    Ruby

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。

  1. 在 Google Cloud 控制台中，进入管理资源页面。

    转到“管理资源”

  2. 在项目列表中，选择要删除的项目，然后点击删除。
  3. 在对话框中输入项目 ID，然后点击关闭以删除项目。