Key value databases come with a high speed and performance where we mostly cannot reach in relational databases. Herein similar to Cassandra, Redis is a fast key value store solution. In this post, we are going to adopt Redis to build an overperforming face recognition application. On the other hand, this could be adapted to NLP studies or any reverse image search case such as in Google Images. The official redis distribution is available for Linux and MacOS here.

OpenCV becomes a de facto standard for image processing studies. The library offers some legacy techniques for face recognition as well. Local binary patterns histograms (LBPH), EigenFace and FisherFace methods are covered in the package. It is a fact that these conventional face recognition algorithms ARE NOT state-of-the-art techniques anymore. Nowadays, CNN based deep learning approaches overperform than these old-fashioned methods.

In a report published Thursday, the American Civil Liberties Union found that Amazon's facial recognition software mistakenly matched 28 U.S. Congresspeople to photos from a mugshot database. The software--which is already in use by some police departments--was disproportionately inaccurate in identifying people of color. In the test, the ACLU used Amazon's Rekognition software to compare photos of the 535 members of the House and Senate to a database of 25,000 mugshots, for an overall inaccuracy rate of 5%. But while only 20% of the members of Congress are non-white, about 40% of the falsely ID'd legislators were men and women of color. The potential outcomes of such misidentifications in life-or-death police encounters are terrifying to consider.

Let's start with some comments about a recent ACLU blog in which they run a facial recognition trial. Using Rekognition, the ACLU built a face database using 25,000 publicly available arrest photos and then performed facial similarity searches of that database using public photos of all current members of Congress. They found 28 incorrect matches out of 535, using an 80% confidence level; this is a 5% misidentification (sometimes called'false positive') rate and a 95% accuracy rate. The ACLU has not published its data set, methodology, or results in detail, so we can only go on what they've publicly said. To illustrate the impact of confidence threshold on false positives, we ran a test where we created a face collection using a dataset of over 850,000 faces commonly used in academia.