Recognition System of Partially Closed Face

A group of researchers introduced a face recognition algorithm that can accurately recognize a person’s partially closed face. It works on the principle of deep learning and recognizes a person by 14 points on his face (corners of the eyes, eyebrows, lips, and so on) necessary for successful computer recognition.

In the process, the algorithm first marks on a person’s photo 14 points he needs on his face using an ultra-accurate neural network. Then the points marked on the hidden face are compared with the points on the fully open face, which are stored in a separate database.

In the course of work on the algorithm, the researchers tested two databases with photographs of disguised faces. The first of them contained photographs of people whose faces were almost completely covered. The second base contained images with slightly covered faces (for example, a cap or sun goggles). During the experiments, the algorithm was able to recognize from the “complex” base with an accuracy of 62%, and from the “simple” one with an accuracy of 78%. Which exceeds the results of the previous version by 9% and 13%, respectively.