Can Rekognition detect object color?

AWS Rekognition does not directly provide the capability to detect the color of objects within an image. Its primary functions include detecting objects, scenes, and activities in images and videos, recognizing faces and their attributes, and identifying text. While it can recognize various attributes and characteristics of objects, such as labels and bounding box coordinates, it doesn't explicitly identify or save the color information of these objects.

For tasks that require color detection, you would typically need to integrate Rekognition with additional image processing libraries or custom algorithms that can analyze pixel values for color information. Libraries such as OpenCV or even Python's PIL (Pillow) can be used to process images on this more granular level, where you can extract and manipulate pixel color data after objects have been identified by Rekognition.

Hi,

As explained previously by Olekssi, AWS Rekognition doesn't label objects with their colors.

But, it recognizes foreground and background colors. So, in some cases, you can achieve want you want: if there is one single object and it is in foreground. You may be able to assign the "dominant foreground color" to this object. But, clearly, it is not as general as what you seem to be looking for.

For all details, see https://aws.amazon.com/blogs/machine-learning/amazon-rekognition-labels-adds-600-new-labels-including-landmarks-and-now-detects-dominant-colors/

Best,

Didier

Instead of Rekognition would using Bedrock / Claude Sonnet be an option in this case? Here's an example:

Prompt: What is the color of the T-shirt of the person taking the selfie?

Response: The person taking the selfie in the image is wearing an orange or yellowish-orange colored t-shirt or top.

Screenshot: