Rekognition

Protocol: JSON 1.1 (X-Amz-Target: RekognitionService.*) Endpoint: POST http://localhost:4566/

Supported Actions

Action	Description
DetectLabels	Detect objects, scenes, and concepts in an image
DetectFaces	Detect faces in an image and their attributes
DetectText	Detect text in an image
CompareFaces	Compare a face in a source image with faces in a target image
DetectModerationLabels	Detect unsafe or inappropriate content in an image

Emulation Behavior

• Stub responses: every action returns a fixed, AWS-shaped response. No image analysis is performed: what an image shows never changes the result.
• Images are checked as Rekognition checks them: an image Rekognition refuses is refused here, with Rekognition's error, before any response. Measured against Rekognition in us-east-2, for each operation and for both Bytes and S3Object:
  • Anything but a JPEG or PNG (WebP, GIF, other bytes) is refused with InvalidImageFormatException, and so is a cut-off or hollow file the way Rekognition judges it: a PNG without an IDAT or IEND chunk, or a JPEG cut inside its headers or whose frame or scan header is shorter than its component count requires. A JPEG whose headers are complete through the start-of-scan header is accepted however much of the image data after it is missing.
  • A JPEG with stray bytes between its segments is accepted, as Rekognition accepts it.
  • Bytes that are not base64 are refused with SerializationException; base64 without its = padding is accepted.
  • Bytes that are empty or longer than 5,242,880 bytes are refused with ValidationException, naming the member (image.bytes, sourceImage.bytes, targetImage.bytes). For CompareFaces both images' lengths are checked before either image is read, and every violation is reported in one error.
  • An S3Object that cannot be read (a missing bucket or object, or a bucket in another region than the request) is refused with InvalidS3ObjectException. A bucket created without a location constraint is in the default region. The object is read from Floci's S3, so it has to be uploaded first; only as much of it as its format needs is read, so a large object is not loaded into memory.
  • A side over 10,000 pixels is refused with ImageTooLargeException by DetectModerationLabels and DetectLabels, and with InvalidImageFormatException by the other operations.
  • Two documented limits are not enforced, because Rekognition does not enforce them: images under 80 pixels are accepted, and an S3 object over 15 MB is accepted.
• Content-listing operations return one canned entry: DetectLabels and DetectText always report a single stub label/text detection, matching the Textract stub's block-hierarchy convention.
• Face/moderation operations return empty results: DetectFaces, CompareFaces, and DetectModerationLabels always report nothing found (empty FaceDetails/FaceMatches/ModerationLabels) rather than fabricating biometric attributes or a moderation flag for content that was never analyzed — an honest "nothing detected" response, matching the Comprehend PII-detection stub's precedent.
• Real input validation: Image (or SourceImage/TargetImage for CompareFaces) is required, must be a JSON structure, and must specify Bytes (string-typed) or S3Object (structure-typed) — matching AWS's own modeling and rejecting wrong JSON types as SerializationException. This is protocol compatibility, not content analysis.
• Both Bytes and S3Object: supplying both is accepted rather than rejected, since nothing in the Rekognition API model declares that combination invalid. The bytes are what is checked.
• Out of scope: face-collection persistence (CreateCollection, IndexFaces, SearchFaces), custom-model training (Projects, Datasets, ProjectVersions), live video (StreamProcessor), async video jobs (Start*/Get* Celebrity/Content/Face/Label/Person/Segment/Text detection), and Face Liveness sessions are not implemented.

Configuration

Variable	Default	Description
FLOCI_SERVICES_REKOGNITION_ENABLED	true	Enable or disable the service
AI_MOCK_CONFIG	unset	Path to a shared mock-response config file — see "Mock Responses" below

Mock Responses

The default stub responses above are the same for every call. To exercise application logic that branches on detection results, point AI_MOCK_CONFIG at a JSON file shared across Rekognition, Textract, and Comprehend:

{
  "rekognition": {
    "my-bucket/cat.jpg": {
      "DetectLabels": {
        "Labels": [{ "Name": "Cat", "Confidence": 97.2 }, { "Name": "Animal", "Confidence": 98.1 }],
        "LabelModelVersion": "1.0"
      }
    }
  }
}

The object named by the key has to exist in Floci's S3 and be an image Rekognition accepts: the image is checked before the mock is looked up, so a call naming an object that was never uploaded fails with InvalidS3ObjectException, as it does on AWS.

The lookup key is "<Bucket>/<Name>" from Image.S3Object (or SourceImage.S3Object for CompareFaces — TargetImage has no independent key in this scheme). A Bytes-backed image has no such key, so mocking only applies to S3Object-based calls. The file is re-read when its modification time changes, so it can be edited without restarting the emulator. A missing file, an unset AI_MOCK_CONFIG, or no matching entry all fall back to the default stub.

Examples

export AWS_ENDPOINT_URL=http://localhost:4566

aws rekognition detect-labels \
  --image '{"S3Object":{"Bucket":"my-bucket","Name":"photo.jpg"}}' \
  --endpoint-url $AWS_ENDPOINT_URL

aws rekognition detect-text \
  --image '{"S3Object":{"Bucket":"my-bucket","Name":"photo.jpg"}}' \
  --endpoint-url $AWS_ENDPOINT_URL

aws rekognition compare-faces \
  --source-image '{"S3Object":{"Bucket":"my-bucket","Name":"source.jpg"}}' \
  --target-image '{"S3Object":{"Bucket":"my-bucket","Name":"target.jpg"}}' \
  --endpoint-url $AWS_ENDPOINT_URL

SDK Example (Java)

RekognitionClient rekognition = RekognitionClient.builder()
    .endpointOverride(URI.create("http://localhost:4566"))
    .region(Region.US_EAST_1)
    .credentialsProvider(StaticCredentialsProvider.create(
        AwsBasicCredentials.create("test", "test")))
    .build();

DetectLabelsResponse response = rekognition.detectLabels(req -> req
    .image(img -> img.s3Object(s3 -> s3.bucket("my-bucket").name("photo.jpg"))));

System.out.println("Labels: " + response.labels());