AWS Rekognition image analysis

With AWS Rekognition, MediaLab analyzes your uploaded images and fills in metadata automatically, so images are easier to find and organize from the moment they are uploaded. It is one of the options for AI metadata detection.

AWS Rekognition is available on request. Contact us to enable it for your account.

What it detects

DetectionWhat it adds
Label detectionObjects, scenes and concepts in the image, such as "car", "beach" or "crowd".
Text detectionText visible in the image, such as signs, captions or product names.
Celebrity detectionWell-known people recognized in the image.
Face detectionFaces in the image and their attributes.
Content moderation detectionPotentially unsafe or inappropriate content, so it can be reviewed before it is published.

The results are saved in metadata fields of your metadata template. Like other metadata, they can be used to search and filter, for example in a smart collection of all images showing a specific product. The complete analysis is also kept with the file.

Choosing which results end up in your metadata

You decide which results are saved by adding fields with these short names to your template in the Metadata Template Editor. Only the fields you add are filled in; you can give them any label you like. Use the field type shown in the table: results that can contain several values, such as labels, need a Select (lookup) field.

Field short nameField typeContents
aws_rekognition_labelsSelect (lookup)Detected labels, most confident first
aws_rekognition_labels_aliasesSelect (lookup)Alternative names for the detected labels
aws_rekognition_labels_categoriesSelect (lookup)Categories of the detected labels, such as "Vehicles and Automotive"
aws_rekognition_text_linesSelect (lookup)Lines of text found in the image
aws_rekognition_celebritiesSelect (lookup)Names of recognized celebrities
aws_rekognition_faces_age_range_low / aws_rekognition_faces_age_range_highNumberEstimated age range of the people in the image
aws_rekognition_faces_gendersSelect (lookup)Estimated genders
aws_rekognition_faces_emotionsSelect (lookup)Detected emotions, such as happy or calm
aws_rekognition_faces_has_smile, aws_rekognition_faces_has_eyes_open, aws_rekognition_faces_has_mouth_open, aws_rekognition_faces_has_sunglassesCheckboxYes/no per face attribute
aws_rekognition_moderation_has_moderationCheckboxYes when potentially unsafe content was found
aws_rekognition_moderation_l1, aws_rekognition_moderation_l2, aws_rekognition_moderation_l3Select (lookup)Moderation labels, from broad (level 1) to specific (level 3)

Detected values never overwrite metadata that was already filled in, so manual input is kept.

Fine-tuning the results

Administrators can adjust the analysis under Settings → Extensions → AWS Rekognition:

  • Switch each type of detection on or off.
  • Minimum confidence: only keep results the AI is sufficiently sure about.
  • Maximum labels and Maximum lines of text: limit how many results are saved per image.
  • Label categories excluded and Labels excluded: leave out labels you don't need, such as entire categories.

Changes apply to images analyzed after saving.

Good to know

  • AWS Rekognition analyzes images. For video and audio, see the enterprise options in AI metadata detection.
  • AI detection is not perfect. Users with edit rights can correct or remove detected values like any other metadata.
  • The analysis runs after upload, so results appear shortly after an image has been processed.

References