Amazon Rekognition launches face occlusion detection to improve identity verification accuracy

Today, AWS announced the general availability of face occlusion detection to improve face verification accuracy. The new FaceOccluded attribute in Amazon Rekognition DetectFaces and IndexFaces APIs detects if the face in an image is partially captured or not fully visible due to overlapping objects, clothing, and body parts.

Financial services, gig economy, telco, healthcare, social media, and other customers use facial verification during online onboarding, step-up authentication, and age-based access restriction. These customers perform quality checks on captured images of faces to reject images with sub-optimal face brightness, sharpness, pose, and size among other attributes. Starting today, these customers can use the FaceOccluded attribute available in DetectFaces and IndexFaces APIs as an additional quality check. FaceOccluded returns "true" with a high confidence score if a detected face’s eyes, nose, or mouth are partially captured or if they are covered by masks, dark sunglasses, cell phones, hands, or other objects. When face occlusion is detected, customers can request an un-occluded face image from users to significantly improve face verification accuracy or support other use cases that require faces to be fully visible. To optimize the end user experience, FaceOccluded returns “false” with a high confidence score if common occurrences that do not impact face verification are detected, such as eye glasses, lightly tinted sunglasses, strands of hair, and others.

Face occlusion detection is now available in all AWS regions supported by Amazon Rekognition Faces. To get started, please download the latest AWS SDK and visit our Amazon Rekognition Faces documentation.