Get keywords for images from the Google Cloud Vision API with C

This blog will explore how to get keywords for images using the Google Cloud Vision API. In order to use this API you'll need a Google Console Project. I'll be using a JSON server-to-server token. If you're not sure how to set this up, please consult the Quickstart.

According to Google the project has a lot to offer:

Google Cloud Vision API enables developers to understand the content of an image by encapsulating powerful machine learning models in an easy to use REST API. It quickly classifies images into thousands of categories (e.g., "sailboat", "lion", "Eiffel Tower"), detects individual objects and faces within images, and finds and reads printed words contained within images. You can build metadata on your image catalog, moderate offensive content, or enable new marketing scenarios through image sentiment analysis. Analyze images uploaded in the request or integrate with your image storage on Google Cloud Storage.

Nuget

Never build yourself what others have been sweating to build for you! Google has provided a fine Nuget package for the API.

dotnet add package Google.Apis.Vision.v1 --version 1.49.0.2142

Credentials + Service

The Google API uses many types of credentials. I'm using the server-to-server credential in a JSON format. The following code will create those credentials. Don't forget to provide the scope, otherwise the credentials will fail!

/// <summary>
/// Creates the credentials.
/// </summary>
/// <param name="path">The path to credential file.</param>
/// <returns>The credentials.</returns>
public static GoogleCredential CreateCredentials(string path)
{
    GoogleCredential credential;
    using (var stream = new FileStream(path, FileMode.Open, FileAccess.Read))
    {
        var c = GoogleCredential.FromStream(stream);
        credential = c.CreateScoped(VisionService.Scope.CloudPlatform);
    }

    return credential;
}

Now we can create the service using the credentials:

/// <summary>
/// Creates the service.
/// </summary>
/// <param name="applicationName">Name of the application.</param>
/// <param name="credentials">The credentials.</param>
/// <returns>The service.</returns>
public static VisionService CreateService(
    string applicationName, 
    IConfigurableHttpClientInitializer credentials)
{
    var service = new VisionService(
        new BaseClientService.Initializer()
        {
            ApplicationName = applicationName,
            HttpClientInitializer = credentials
        }
    );

    return service;
}

Features

The Cloud Vision API returns keywords based on features. You can use the following feature constants:

Prepare image request

First, let's create an AnnotateImageRequest that represent the data for a single file.

/// <summary>
/// Creates the annotation image request.
/// </summary>
/// <param name="path">The path.</param>
/// <param name="featureTypes">The feature types.</param>
/// <returns>The request.</returns>
private static AnnotateImageRequest CreateAnnotationImageRequest(
    string path, 
    string[] featureTypes)
{
    if (!File.Exists(path))
    {
        throw new FileNotFoundException("Not found.", path);
    }

    var request = new AnnotateImageRequest();
    request.Image = new Image();

    var bytes = File.ReadAllBytes(path);
    request.Image.Content = Convert.ToBase64String(bytes);

    request.Features = new List<Feature>();

    foreach(var featureType in featureTypes)
    {
        request.Features.Add(new Feature() { Type = featureType });
    }

    return request;
}

Note that you'll need to new up any property you need to use on the AnnotationImageRequest. The API will not create collections like Features.

AnnotateAsync - single file

Now we can create an extension method that will extend the VisionService with a simple method to execute an annotation request for a single file. It ties the service and the AnnotateImageRequest together.

/// <summary>
/// Annotates the file asynchronously.
/// </summary>
/// <param name="service">The service.</param>
/// <param name="file">The file.</param>
/// <param name="features">The features.</param>
/// <returns>The annotation response.</returns>
public static async Task<AnnotateImageResponse> AnnotateAsync(
    this VisionService service, 
    FileInfo file, 
    params string[] features)
{
    var request = new BatchAnnotateImagesRequest();
    request.Requests = new List<AnnotateImageRequest>();
    request.Requests.Add(CreateAnnotationImageRequest(file.FullName, features));

    var result = await service.Images.Annotate(request).ExecuteAsync();

    if (result?.Responses?.Count > 0)
    {
        return result.Responses[0];
    }

    return null;
}

The API supports batched requests. Implementation should not be hard, you'll just need to add more requests (request.Request.Add). I left it out of this tutorial.

Proof of concept

All the components are ready. I've created a directory with some files from the excellent Unsplash project and I ran the following program:

//find the files
var ext = new HashSet<string>(StringComparer.OrdinalIgnoreCase) { ".png", ".jpg", ".gif" };
var dir = @"C:\temp\Examples";
var files =
    Directory
        .GetFiles(dir, "*.*", SearchOption.AllDirectories)
        .Where(f => ext.Contains(Path.GetExtension(f)))
        .Select(f => new FileInfo(f))
        .ToArray();

//create service
var credentails = GoogleVisionApi.CreateCredentials("user_credentials.json");
var service = GoogleVisionApi.CreateService("MyApplication", credentails);

//process each file
foreach (var file in files)
{
    string f = file.FullName;
    Console.WriteLine("Reading " + f + ":");

    var task = service.AnnotateAsync(file, "LABEL_DETECTION");
    var result = task.Result;

    var keywords = result?.LabelAnnotations?.Select(s => s.Description).ToArray();
    var words = String.Join(", ", keywords);
    Console.WriteLine(words);

    f += ".keywords.txt";
    File.WriteAllText(f, words);
}

Results

This will produce the following results:

.thumbnail-results td { width:33%; }

Final thoughts

The keywords are fine, but I had expected more from the Cloud Vision API. I'm missing keywords like: sun, puppy, strawberry, pots, window-sill. Maybe the service will improve. I can't wait to experiment with other services. Hopefully the code will help your project.