You work in the security department of your company's IT division.
Your company has decided to try to use facial recognition to improve security on their campus.
You have been asked to design a system that augments your company's building access security by scanning the faces of people entering their buildings and recognizing the person as either an employee/contractor/consultant, who is in the company's database, or visitor, who is not in their database. Your company has over 750,000 employees and over 250,000 contractors and consultants across their many campus locations worldwide.
These workers are all registered in their HR database.
Each of these workers has an image of their face stored in the HR database.
You have decided to use Amazon Rekognition for your facial recognition solution.
On occasion, the Rekognition model fails to recognize visitors to the buildings.
What could be the source of the problem?
Click on the arrows to vote for the correct answer
A. B. C. D.Answer: D.
Option A is incorrect.
From the Amazon Rekognition FAQs: “Face landmarks are a set of salient points, usually located on the corners, tips or midpoints of key facial components such as the eyes, nose, and mouth.
Amazon Rekognition DetectFaces API returns a set of face landmarks that can be used to crop faces, morph one face into another, overlay custom masks to create custom filters, and more.” Face landmarks don't have a sharpness parameter.
Option B is incorrect.
The bounding box and confidence score are used to determine confidence in the Rekognition comparison result.
A maximum confidence score tolerance would not cause failures to recognize faces.
A low confidence score tolerance would do.
Option C is incorrect.
Similar to option C, the default threshold would not be a common source of recognition failure.
A confidence threshold tolerance that is set too low would cause a failure in recognition.
Option D is correct.
A suboptimal face collection can be the source of recognition failure.
Our face collection has only one image per person.
The recommendation from the Amazon Rekognition FAQs is “Besides video resolution, the quality and representative faces part of the face collections to search has a major impact.
Using multiple face instances per person with variations like beard, glasses, poses (profile and frontal) will significantly improve the performance.”
Reference:
Please see the Amazon Recognition developer guide titled Detecting and Analyzing Faces.
The source of the problem could be related to multiple factors, but one possible explanation is related to the confidence threshold tolerance setting in Amazon Rekognition.
Amazon Rekognition is a machine learning-based image and video analysis service that provides facial recognition capabilities. When it performs facial recognition, it analyzes the input image and compares it to a set of reference images (known as a face collection) to identify the individual in the input image.
To identify a person in an image, Rekognition calculates a confidence score, which is a numerical value between 0 and 100 that indicates the level of confidence that the model has in the match. The higher the confidence score, the more likely the match is to be accurate.
By default, Rekognition sets a confidence threshold of 80%, which means that it will only consider matches with a confidence score of 80 or above to be a positive match. If the confidence score is lower than the threshold, the model will not recognize the face.
In the case of visitors, their faces may not be present in the company's HR database, which means that Rekognition will not be able to identify them as employees/contractors/consultants. As a result, Rekognition will attempt to identify them as visitors by comparing their faces to the faces in the face collection of visitors.
However, if the confidence threshold is set too high, Rekognition may fail to recognize some visitors, even if their faces are present in the face collection. For example, if the threshold is set to 95%, Rekognition will only consider matches with a confidence score of 95 or above to be a positive match. If the visitor's face is similar but not identical to a face in the face collection, the confidence score may be lower than 95, causing Rekognition to fail to recognize them.
Therefore, the source of the problem could be that the confidence threshold tolerance is set to the default value or to a value that is too high, which causes Rekognition to miss some matches and fail to recognize visitors. To solve this problem, the confidence threshold tolerance may need to be lowered to improve the model's ability to recognize visitors.