Menggunakan Cloud Vision API untuk menentukan apakah gambar aman

Tutorial ini menunjukkan penggunaan Cloud Run, Cloud Vision API, dan ImageMagick untuk mendeteksi dan memburamkan gambar menyinggung yang diupload ke bucket Cloud Storage.

Mempelajari lebih lanjut

Untuk dokumentasi mendetail yang menyertakan contoh kode ini, lihat artikel berikut:

• Tutorial memproses gambar dari Cloud Storage
• Memproses gambar secara asinkron

Contoh kode

Go

Untuk melakukan autentikasi ke Cloud Run, siapkan Kredensial Default Aplikasi. Untuk mengetahui informasi selengkapnya, lihat Menyiapkan autentikasi untuk lingkungan pengembangan lokal.

// GCSEvent is the payload of a GCS event.
type GCSEvent struct {
	Bucket string `json:"bucket"`
	Name   string `json:"name"`
}

// BlurOffensiveImages blurs offensive images uploaded to GCS.
func BlurOffensiveImages(ctx context.Context, e GCSEvent) error {
	outputBucket := os.Getenv("BLURRED_BUCKET_NAME")
	if outputBucket == "" {
		return errors.New("BLURRED_BUCKET_NAME must be set")
	}

	img := vision.NewImageFromURI(fmt.Sprintf("gs://%s/%s", e.Bucket, e.Name))

	resp, err := visionClient.DetectSafeSearch(ctx, img, nil)
	if err != nil {
		return fmt.Errorf("AnnotateImage: %w", err)
	}

	if resp.GetAdult() == visionpb.Likelihood_VERY_LIKELY ||
		resp.GetViolence() == visionpb.Likelihood_VERY_LIKELY {
		return blur(ctx, e.Bucket, outputBucket, e.Name)
	}
	log.Printf("The image %q was detected as OK.", e.Name)
	return nil
}

Java

Untuk melakukan autentikasi ke Cloud Run, siapkan Kredensial Default Aplikasi. Untuk mengetahui informasi selengkapnya, lihat Menyiapkan autentikasi untuk lingkungan pengembangan lokal.

// Blurs uploaded images that are flagged as Adult or Violence.
public static void blurOffensiveImages(JsonObject data) {
  String fileName = data.get("name").getAsString();
  String bucketName = data.get("bucket").getAsString();
  BlobInfo blobInfo = BlobInfo.newBuilder(bucketName, fileName).build();
  // Construct URI to GCS bucket and file.
  String gcsPath = String.format("gs://%s/%s", bucketName, fileName);
  System.out.println(String.format("Analyzing %s", fileName));

  // Construct request.
  List<AnnotateImageRequest> requests = new ArrayList<>();
  ImageSource imgSource = ImageSource.newBuilder().setImageUri(gcsPath).build();
  Image img = Image.newBuilder().setSource(imgSource).build();
  Feature feature = Feature.newBuilder().setType(Type.SAFE_SEARCH_DETECTION).build();
  AnnotateImageRequest request =
      AnnotateImageRequest.newBuilder().addFeatures(feature).setImage(img).build();
  requests.add(request);

  // Send request to the Vision API.
  try (ImageAnnotatorClient client = ImageAnnotatorClient.create()) {
    BatchAnnotateImagesResponse response = client.batchAnnotateImages(requests);
    List<AnnotateImageResponse> responses = response.getResponsesList();
    for (AnnotateImageResponse res : responses) {
      if (res.hasError()) {
        System.out.println(String.format("Error: %s\n", res.getError().getMessage()));
        return;
      }
      // Get Safe Search Annotations
      SafeSearchAnnotation annotation = res.getSafeSearchAnnotation();
      if (annotation.getAdultValue() == 5 || annotation.getViolenceValue() == 5) {
        System.out.println(String.format("Detected %s as inappropriate.", fileName));
        blur(blobInfo);
      } else {
        System.out.println(String.format("Detected %s as OK.", fileName));
      }
    }
  } catch (Exception e) {
    System.out.println(String.format("Error with Vision API: %s", e.getMessage()));
  }
}

Node.js

Untuk melakukan autentikasi ke Cloud Run, siapkan Kredensial Default Aplikasi. Untuk mengetahui informasi selengkapnya, lihat Menyiapkan autentikasi untuk lingkungan pengembangan lokal.

// Blurs uploaded images that are flagged as Adult or Violence.
exports.blurOffensiveImages = async event => {
  // This event represents the triggering Cloud Storage object.
  const object = event;

  const file = storage.bucket(object.bucket).file(object.name);
  const filePath = `gs://${object.bucket}/${object.name}`;

  console.log(`Analyzing ${file.name}.`);

  try {
    const [result] = await client.safeSearchDetection(filePath);
    const detections = result.safeSearchAnnotation || {};

    if (
      // Levels are defined in https://cloud.google.com/vision/docs/reference/rest/v1/AnnotateImageResponse#likelihood
      detections.adult === 'VERY_LIKELY' ||
      detections.violence === 'VERY_LIKELY'
    ) {
      console.log(`Detected ${file.name} as inappropriate.`);
      return blurImage(file, BLURRED_BUCKET_NAME);
    } else {
      console.log(`Detected ${file.name} as OK.`);
    }
  } catch (err) {
    console.error(`Failed to analyze ${file.name}.`, err);
    throw err;
  }
};

Python

Untuk melakukan autentikasi ke Cloud Run, siapkan Kredensial Default Aplikasi. Untuk mengetahui informasi selengkapnya, lihat Menyiapkan autentikasi untuk lingkungan pengembangan lokal.

def blur_offensive_images(data):
    """Blurs uploaded images that are flagged as Adult or Violence.

    Args:
        data: Pub/Sub message data
    """
    file_data = data

    file_name = file_data["name"]
    bucket_name = file_data["bucket"]

    blob = storage_client.bucket(bucket_name).get_blob(file_name)
    blob_uri = f"gs://{bucket_name}/{file_name}"
    blob_source = vision.Image(source=vision.ImageSource(image_uri=blob_uri))

    # Ignore already-blurred files
    if file_name.startswith("blurred-"):
        print(f"The image {file_name} is already blurred.")
        return

    print(f"Analyzing {file_name}.")

    result = vision_client.safe_search_detection(image=blob_source)
    detected = result.safe_search_annotation

    # Process image
    if detected.adult == 5 or detected.violence == 5:
        print(f"The image {file_name} was detected as inappropriate.")
        return __blur_image(blob)
    else:
        print(f"The image {file_name} was detected as OK.")

Langkah berikutnya

Untuk menelusuri dan memfilter contoh kode untuk produk Google Cloud lainnya, lihat Google Cloud browser contoh.