Face Biometrics as a Potential Predictor for COVID-19 Susceptibility

Fighting against the COVID-19 pandemic caused by the SARS-CoV-2 virus is one of the most critical challenges facing the global health system today. The possibility to identify the group of persons in the cohort of people under 50 years old, who are sensitive to the COVID-disease by non-invasive methods, is a very perspective approach for estimating the epidemiological state of the human population. The study aimed to identify the features of people's faces with COVID-19 that the most correlate with disease severity could serve as one of these approaches. For this aim, 525 photos of patients' faces with different outcomes of COVID-19 disease were analyzed using the Dlib face recognition convolutional neural network pre-trained for face recognition. Face descriptor vectors were obtained using the convolutional neural network. Facial features were found that predict a person's sensitivity to the SARS-CoV-2 virus (disease severity), and the contribution of each of the features to the risk of developing a severe form of COVID in a person was found. The accuracy of the binary classification of the individual severity of the COVID-19 course using the k-nearest neighbors algorithm on the test dataset was accuracy - 84%, AUC - 0.90.

Hybrid Technique on Image Clustering

Face recognition using FLD for extracting high dimensional images is introduced in this paper. The main purpose is to work on removing bugs and noise from the images and extract the facial expression applied on face descriptor. FLD is selected for increasing the discrimination information [17]. The main points of this paper give the brief knowledge about the face recognition and face clustering. Its shows how biometric terms help the local and global features for extracting information from database. Finding better solutions to deal with noise in face recognition is a challenging task [18]. We also performed some comparative analysis on various face recognition techniques. The main motive of this paper is to increase the recognition rate of the images and provide good efficiency. This method defines how the features and facial expression are extracted and all noise and bugs are eliminated to make a separate individual cluster of same known faces.

A New Local Descriptor Based on Strings for Face Recognition

This paper proposes the use of strings as a new local descriptor for face recognition. The face image is first divided into nonoverlapping subregions from which the strings (words) are extracted using the principle of chain code algorithm and assigned into the nearest words in a dictionary of visual words (DoVW) with the Levenshtein distance (LD) by applying the bag of visual words (BoVW) paradigm. As a result, each region is represented by a histogram of dictionary words. The histograms are then assembled as a face descriptor. Our methodology depends on the path pursued from a starting pixel and do not require a model as the other approaches from the literature. Therefore, the information of the local and global properties of an object is obtained. The recognition is performed by using the nearest neighbor classifier with the Hellinger distance (HD) as a comparison between feature vectors. The experimental results on the ORL and Yale databases demonstrate the efficiency of the proposed approach in terms of preserving information and recognition rate compared to the existing face recognition methods.