Facial recognition accumulates problems

Facial recognition accumulates problems

An NBC investigation reveals the use of images without consent

  Facial recognition continues to make inroads in the world of smartphones and also in the different security devices in cities. However, although it keeps updating and improving its algorithm, it shows more and more bias problems.

The most advanced in the field is Amazon with its Rekognition system, although it has been involved in more than one scandal. Now, the ball is on the roof of IBM after an investigation of the US network NBC.

   Algorithms and machine learning need continuous training with images and the large bank of photographs is internet. The network of networks stores a wide range of faces. Users upload their favorite photos to the cloud, share their funniest moments on social networks.

With this information, researchers and academics can collect hundreds of data on sex, age, race and facial features, despite all the final problems of facial recognition tools.

This investigation, revealed by NBC, points to IBM as the latest company to join this training. The American technology launched in January a collection of almost one million photos taken from Flickr, according to NBC.

Through these images, IBM worked to reduce the bias present in facial recognition tools. However, the research reveals that the technology company did not contact the users and owners of the images for the company's investigation.

For training the IBM algorithm they took the images with details that included facial geometry and skin tone and that could be used to develop facial recognition algorithms.

Before the start-up of this type of tools, the software needs a set-up for its correct operation. Beyond its final goal, this type of tool works in four steps.

Before activating, the cameras that make up the system capture the face of the person to be identified in the chosen device. Next, the program obtains the biometric information of the facial features, which make up the so-called facial biometric pattern.

Later, the biometric information compares the information gathered with the existing ones in the database and this is where the biggest problems have been found in Amazon and where IBM has built its large gallery from Flickr.

Finally, the facial recognition system identifies the face as the one that has obtained the highest percentage of similarity among those consulted, based on a very high coincidence threshold.

According to a study conducted by the MIT Media Lab, facial recognition is more accurate for white men and has many more flaws for people with darker skin, especially women. Specifically, this report noted that Microsoft and IBM applications erroneously identified women with darker skin 35% of the time and men with darker skin 12%.