ImageMagick のチュートリアル(第 1 世代)

このチュートリアルでは、Cloud Run functions、Cloud Vision APIImageMagick を使用して、Cloud Storage バケットにアップロードされた不適切な画像を検出してぼかす方法を説明します。

データの流れ

ImageMagick チュートリアル アプリケーションでは、データの流れは次のようになります。

  1. 画像が Cloud Storage バケットにアップロードされます。
  2. 関数が Vision API を使用して画像を分析します。
  3. 暴力的なコンテンツやアダルト コンテンツが見つかった場合、関数は ImageMagick を使用して画像をぼかします。
  4. ぼかしの入った画像が別の Cloud Storage バケットにアップロードされます。

アプリケーションの準備

  1. 画像をアップロードする Cloud Storage バケットを作成します。YOUR_INPUT_BUCKET_NAME は、グローバルに固有のバケット名です。

    gcloud storage buckets create gs://YOUR_INPUT_BUCKET_NAME
  2. ぼかし入りの画像を保存する Cloud Storage バケットを作成します。YOUR_OUTPUT_BUCKET_NAME はグローバルに固有のバケット名です。

    gcloud storage buckets create gs://YOUR_OUTPUT_BUCKET_NAME
  3. ローカルマシンにサンプルアプリのリポジトリのクローンを作成します。

    Node.js

    git clone https://github.com/GoogleCloudPlatform/nodejs-docs-samples.git

    または、zip 形式のサンプルをダウンロードし、ファイルを抽出してもかまいません。

    Python

    git clone https://github.com/GoogleCloudPlatform/python-docs-samples.git

    または、zip 形式のサンプルをダウンロードし、ファイルを抽出してもかまいません。

    Go

    git clone https://github.com/GoogleCloudPlatform/golang-samples.git

    または、zip 形式のサンプルをダウンロードし、ファイルを抽出してもかまいません。

    Java

    git clone https://github.com/GoogleCloudPlatform/java-docs-samples.git

    または、zip 形式のサンプルをダウンロードし、ファイルを抽出してもかまいません。

    Ruby

    git clone https://github.com/GoogleCloudPlatform/ruby-docs-samples.git

    または、zip 形式のサンプルをダウンロードしてファイルを抽出してもかまいません。

  4. Cloud Run functions のサンプルコードが含まれているディレクトリに移動します。

    Node.js

    cd nodejs-docs-samples/functions/imagemagick/

    Python

    cd python-docs-samples/functions/imagemagick/

    Go

    cd golang-samples/functions/imagemagick/

    Java

    cd java-docs-samples/functions/imagemagick/

    Ruby

    cd ruby-docs-samples/functions/imagemagick/

コードについて

依存関係をインポートする

アプリケーションがGoogle Cloud サービス、ImageMagick、ファイル システムを利用するには、複数の依存関係をインポートする必要があります。

Node.js

const gm = require('gm').subClass({imageMagick: true});
const fs = require('fs').promises;
const path = require('path');
const vision = require('@google-cloud/vision');

const {Storage} = require('@google-cloud/storage');
const storage = new Storage();
const client = new vision.ImageAnnotatorClient();

const {BLURRED_BUCKET_NAME} = process.env;

Python

import os
import tempfile

from google.cloud import storage, vision
from wand.image import Image

storage_client = storage.Client()
vision_client = vision.ImageAnnotatorClient()

Go


// Package imagemagick contains an example of using ImageMagick to process a
// file uploaded to Cloud Storage.
package imagemagick

import (
	"context"
	"errors"
	"fmt"
	"log"
	"os"
	"os/exec"

	"cloud.google.com/go/storage"
	vision "cloud.google.com/go/vision/apiv1"
	"cloud.google.com/go/vision/v2/apiv1/visionpb"
)

// Global API clients used across function invocations.
var (
	storageClient *storage.Client
	visionClient  *vision.ImageAnnotatorClient
)

func init() {
	// Declare a separate err variable to avoid shadowing the client variables.
	var err error

	storageClient, err = storage.NewClient(context.Background())
	if err != nil {
		log.Fatalf("storage.NewClient: %v", err)
	}

	visionClient, err = vision.NewImageAnnotatorClient(context.Background())
	if err != nil {
		log.Fatalf("vision.NewAnnotatorClient: %v", err)
	}
}

Java



import com.google.cloud.functions.BackgroundFunction;
import com.google.cloud.functions.Context;
import com.google.cloud.storage.Blob;
import com.google.cloud.storage.BlobId;
import com.google.cloud.storage.BlobInfo;
import com.google.cloud.storage.Storage;
import com.google.cloud.storage.StorageOptions;
import com.google.cloud.vision.v1.AnnotateImageRequest;
import com.google.cloud.vision.v1.AnnotateImageResponse;
import com.google.cloud.vision.v1.BatchAnnotateImagesResponse;
import com.google.cloud.vision.v1.Feature;
import com.google.cloud.vision.v1.Feature.Type;
import com.google.cloud.vision.v1.Image;
import com.google.cloud.vision.v1.ImageAnnotatorClient;
import com.google.cloud.vision.v1.ImageSource;
import com.google.cloud.vision.v1.SafeSearchAnnotation;
import functions.eventpojos.GcsEvent;
import java.io.IOException;
import java.nio.file.Files;
import java.nio.file.Path;
import java.nio.file.Paths;
import java.util.List;
import java.util.logging.Level;
import java.util.logging.Logger;

public class ImageMagick implements BackgroundFunction<GcsEvent> {

  private static Storage storage = StorageOptions.getDefaultInstance().getService();
  private static final String BLURRED_BUCKET_NAME = System.getenv("BLURRED_BUCKET_NAME");
  private static final Logger logger = Logger.getLogger(ImageMagick.class.getName());
}

Ruby

require "functions_framework"

FunctionsFramework.on_startup do
  set_global :storage_client do
    require "google/cloud/storage"
    Google::Cloud::Storage.new
  end

  set_global :vision_client do
    require "google/cloud/vision"
    Google::Cloud::Vision.image_annotator
  end
end

画像の分析

画像の保存用に作成された Cloud Storage バケットに画像がアップロードされると、次の関数が呼び出されます。この関数は、アプロードされた画像を Vision API で分析し、暴力的なコンテンツやアダルト コンテンツを検出します。

Node.js

// 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 await 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

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

Go


// 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

@Override
// Blurs uploaded images that are flagged as Adult or Violence.
public void accept(GcsEvent event, Context context) {
  // Validate parameters
  if (event.getBucket() == null || event.getName() == null) {
    logger.severe("Error: Malformed GCS event.");
    return;
  }

  BlobInfo blobInfo = BlobInfo.newBuilder(event.getBucket(), event.getName()).build();

  // Construct URI to GCS bucket and file.
  String gcsPath = String.format("gs://%s/%s", event.getBucket(), event.getName());
  logger.info(String.format("Analyzing %s", event.getName()));

  // Construct request.
  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();
  List<AnnotateImageRequest> requests = List.of(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()) {
        logger.info(String.format("Error: %s", res.getError().getMessage()));
        return;
      }
      // Get Safe Search Annotations
      SafeSearchAnnotation annotation = res.getSafeSearchAnnotation();
      if (annotation.getAdultValue() == 5 || annotation.getViolenceValue() == 5) {
        logger.info(String.format("Detected %s as inappropriate.", event.getName()));
        blur(blobInfo);
      } else {
        logger.info(String.format("Detected %s as OK.", event.getName()));
      }
    }
  } catch (IOException e) {
    logger.log(Level.SEVERE, "Error with Vision API: " + e.getMessage(), e);
  }
}

Ruby

# Blurs uploaded images that are flagged as Adult or Violence.
FunctionsFramework.cloud_event "blur_offensive_images" do |event|
  # Event-triggered Ruby functions receive a CloudEvents::Event::V1 object.
  # See https://cloudevents.github.io/sdk-ruby/latest/CloudEvents/Event/V1.html
  # The storage event payload can be obtained from the event data.
  payload = event.data
  file_name = payload["name"]
  bucket_name = payload["bucket"]

  # Ignore already-blurred files
  if file_name.start_with? "blurred-"
    logger.info "The image #{file_name} is already blurred."
    return
  end

  # Get image annotations from the Vision service
  logger.info "Analyzing #{file_name}."
  gs_uri = "gs://#{bucket_name}/#{file_name}"
  result = global(:vision_client).safe_search_detection image: gs_uri
  annotation = result.responses.first.safe_search_annotation

  # Respond to annotations by possibly blurring the image
  if annotation.adult == :VERY_LIKELY || annotation.violence == :VERY_LIKELY
    logger.info "The image #{file_name} was detected as inappropriate."
    blur_image bucket_name, file_name
  else
    logger.info "The image #{file_name} was detected as OK."
  end
end

画像のぼかし

アップロードされた画像で暴力的なコンテンツやアダルト コンテンツが見つかると、次の関数が呼び出されます。この関数は不適切な画像をダウンロードして、ImageMagick で画像をぼかし、修正後の画像をアップロードして元の画像を上書きします。

Node.js

// Blurs the given file using ImageMagick, and uploads it to another bucket.
const blurImage = async (file, blurredBucketName) => {
  const tempLocalPath = `/tmp/${path.parse(file.name).base}`;

  // Download file from bucket.
  try {
    await file.download({destination: tempLocalPath});

    console.log(`Downloaded ${file.name} to ${tempLocalPath}.`);
  } catch (err) {
    throw new Error(`File download failed: ${err}`);
  }

  await new Promise((resolve, reject) => {
    gm(tempLocalPath)
      .blur(0, 16)
      .write(tempLocalPath, (err, stdout) => {
        if (err) {
          console.error('Failed to blur image.', err);
          reject(err);
        } else {
          console.log(`Blurred image: ${file.name}`);
          resolve(stdout);
        }
      });
  });

  // Upload result to a different bucket, to avoid re-triggering this function.
  const blurredBucket = storage.bucket(blurredBucketName);

  // Upload the Blurred image back into the bucket.
  const gcsPath = `gs://${blurredBucketName}/${file.name}`;
  try {
    await blurredBucket.upload(tempLocalPath, {destination: file.name});
    console.log(`Uploaded blurred image to: ${gcsPath}`);
  } catch (err) {
    throw new Error(`Unable to upload blurred image to ${gcsPath}: ${err}`);
  }

  // Delete the temporary file.
  return fs.unlink(tempLocalPath);
};

Python

# Blurs the given file using ImageMagick.
def __blur_image(current_blob):
    file_name = current_blob.name
    _, temp_local_filename = tempfile.mkstemp()

    # Download file from bucket.
    current_blob.download_to_filename(temp_local_filename)
    print(f"Image {file_name} was downloaded to {temp_local_filename}.")

    # Blur the image using ImageMagick.
    with Image(filename=temp_local_filename) as image:
        image.blur(radius=0, sigma=16)
        image.save(filename=temp_local_filename)

    print(f"Image {file_name} was blurred.")

    # Upload result to a second bucket, to avoid re-triggering the function.
    # You could instead re-upload it to the same bucket + tell your function
    # to ignore files marked as blurred (e.g. those with a "blurred" prefix)
    blur_bucket_name = os.getenv("BLURRED_BUCKET_NAME")
    blur_bucket = storage_client.bucket(blur_bucket_name)
    new_blob = blur_bucket.blob(file_name)
    new_blob.upload_from_filename(temp_local_filename)
    print(f"Blurred image uploaded to: gs://{blur_bucket_name}/{file_name}")

    # Delete the temporary file.
    os.remove(temp_local_filename)

Go


// blur blurs the image stored at gs://inputBucket/name and stores the result in
// gs://outputBucket/name.
func blur(ctx context.Context, inputBucket, outputBucket, name string) error {
	inputBlob := storageClient.Bucket(inputBucket).Object(name)
	r, err := inputBlob.NewReader(ctx)
	if err != nil {
		return fmt.Errorf("NewReader: %w", err)
	}

	outputBlob := storageClient.Bucket(outputBucket).Object(name)
	w := outputBlob.NewWriter(ctx)
	defer w.Close()

	// Use - as input and output to use stdin and stdout.
	cmd := exec.Command("convert", "-", "-blur", "0x8", "-")
	cmd.Stdin = r
	cmd.Stdout = w

	if err := cmd.Run(); err != nil {
		return fmt.Errorf("cmd.Run: %w", err)
	}

	log.Printf("Blurred image uploaded to gs://%s/%s", outputBlob.BucketName(), outputBlob.ObjectName())

	return nil
}

Java

// Blurs the file described by blobInfo using ImageMagick,
// and uploads it to the blurred bucket.
private static void blur(BlobInfo blobInfo) throws IOException {
  String bucketName = blobInfo.getBucket();
  String fileName = blobInfo.getName();

  // Download image
  Blob blob = storage.get(BlobId.of(bucketName, fileName));
  Path download = Paths.get("/tmp/", fileName);
  blob.downloadTo(download);

  // Construct the command.
  Path upload = Paths.get("/tmp/", "blurred-" + fileName);
  List<String> args = List.of("convert", download.toString(), "-blur", "0x8", upload.toString());
  try {
    ProcessBuilder pb = new ProcessBuilder(args);
    Process process = pb.start();
    process.waitFor();
  } catch (Exception e) {
    logger.info(String.format("Error: %s", e.getMessage()));
  }

  // Upload image to blurred bucket.
  BlobId blurredBlobId = BlobId.of(BLURRED_BUCKET_NAME, fileName);
  BlobInfo blurredBlobInfo =
      BlobInfo.newBuilder(blurredBlobId).setContentType(blob.getContentType()).build();

  byte[] blurredFile = Files.readAllBytes(upload);
  storage.create(blurredBlobInfo, blurredFile);
  logger.info(
      String.format("Blurred image uploaded to: gs://%s/%s", BLURRED_BUCKET_NAME, fileName));

  // Remove images from fileSystem
  Files.delete(download);
  Files.delete(upload);
}

Ruby

require "tempfile"
require "mini_magick"

# Blurs the given file using ImageMagick.
def blur_image bucket_name, file_name
  tempfile = Tempfile.new
  begin
    # Download the image file
    bucket = global(:storage_client).bucket bucket_name
    file = bucket.file file_name
    file.download tempfile
    tempfile.close

    # Blur the image using ImageMagick
    MiniMagick::Image.new tempfile.path do |image|
      image.blur "0x16"
    end
    logger.info "Image #{file_name} was blurred"

    # Upload result to a second bucket, to avoid re-triggering the function.
    # You could instead re-upload it to the same bucket and tell your function
    # to ignore files marked as blurred (e.g. those with a "blurred" prefix.)
    blur_bucket_name = ENV["BLURRED_BUCKET_NAME"]
    blur_bucket = global(:storage_client).bucket blur_bucket_name
    blur_bucket.create_file tempfile.path, file_name
    logger.info "Blurred image uploaded to gs://#{blur_bucket_name}/#{file_name}"
  ensure
    # Ruby will remove the temp file when garbage collecting the object,
    # but it is good practice to remove it explicitly.
    tempfile.unlink
  end
end

関数のデプロイ

ストレージ トリガーを使用して関数をデプロイするには、サンプルコード(Java の場合は pom.xml ファイル)を含むディレクトリで次のコマンドを実行します。

Node.js

gcloud functions deploy blurOffensiveImages \
--no-gen2 \
--runtime=RUNTIME \
--trigger-bucket=YOUR_INPUT_BUCKET_NAME \
--set-env-vars=BLURRED_BUCKET_NAME=YOUR_OUTPUT_BUCKET_NAME

Python

gcloud functions deploy blur_offensive_images \
--no-gen2 \
--runtime=RUNTIME \
--trigger-bucket=YOUR_INPUT_BUCKET_NAME \
--set-env-vars=BLURRED_BUCKET_NAME=YOUR_OUTPUT_BUCKET_NAME

Go

gcloud functions deploy BlurOffensiveImages \
--no-gen2 \
--runtime=RUNTIME \
--trigger-bucket=YOUR_INPUT_BUCKET_NAME \
--set-env-vars=BLURRED_BUCKET_NAME=YOUR_OUTPUT_BUCKET_NAME

Java

gcloud functions deploy java-blur-function \
--no-gen2 \
--entry-point=functions.ImageMagick \
--runtime=RUNTIME \
--memory 512MB \
--trigger-bucket=YOUR_INPUT_BUCKET_NAME \
--set-env-vars=BLURRED_BUCKET_NAME=YOUR_OUTPUT_BUCKET_NAME

C#

gcloud functions deploy csharp-blur-function \
--no-gen2 \
--entry-point=ImageMagick.Function \
--runtime=RUNTIME \
--trigger-bucket=YOUR_INPUT_BUCKET_NAME \
--set-env-vars=BLURRED_BUCKET_NAME=YOUR_OUTPUT_BUCKET_NAME

Ruby

gcloud functions deploy blur_offensive_images \
--no-gen2 \
--runtime=RUNTIME \
--trigger-bucket=YOUR_INPUT_BUCKET_NAME \
--set-env-vars=BLURRED_BUCKET_NAME=YOUR_OUTPUT_BUCKET_NAME

次のように置き換えます。

  • RUNTIME: Ubuntu 18.04 に基づくランタイム(以降のランタイムには、ImageMagick のサポートは含まれていません)。
  • YOUR_INPUT_BUCKET_NAME: 画像をアップロードする Cloud Storage バケットの名前。
  • YOUR_OUTPUT_BUCKET_NAME: ぼかしの入った画像を保存するバケットの名前。

この特定の例では、deploy コマンドのバケット名に gs:// を含めないでください。

画像のアップロード

  1. 肉食ゾンビの画像など、不適切な画像をアップロードします。

    gcloud storage cp zombie.jpg gs://YOUR_INPUT_BUCKET_NAME

    YOUR_INPUT_BUCKET_NAME は、以前に画像のアップロード用に作成した Cloud Storage バケットです。

  2. 実行した内容が完了していることをログで確認します。

    gcloud functions logs read --limit 100
  3. 以前に作成した YOUR_OUTPUT_BUCKET_NAME Cloud Storage バケットで、ぼかし入りの画像を確認できます。