![Scientific and technical journal «Priborostroenie»](/images/mag-pr.png)
APPLICATION OF TRAINING DATA SYNTHESIS METHODS FOR RECOGNITION OF PARTIALLY HIDDEN FACES IN IMAGES
![Scientific and technical journal «Priborostroenie»](/images/mag-pr.png)
Annotation
A new approach to solving the problem of automatic face recognition of people using personal protective equipment such as a medical mask has been proposed and tested. This approach is based on the use of methods of generating synthetic images of partially hidden faces and the face recognition model ArcFace. A strategy for training data sets formation is proposed and a number of corresponding recognition models are derived. A series of experiments aimed at assessing the quality of predictions of the obtained solution are carried out, and a relationship between the resulting quality of predictions implemented by recognition models and the volume of synthetic images in training datasets is established. According to the results of experimental studies, neural network models, further trained on datasets with volume of artificially synthesized images of 40-60%, demonstrate values of recognition accuracy above 87% on the AAc quantitative metric (Average Accuracy). Using the proposed approach makes it possible to significantly improve the quality of recognition of partially hidden faces compared to the basic approach.
Keywords
Постоянный URL
Articles in current issue
- METHODOLOGICAL AND METHODICAL PRINCIPLES OF PROACTIVE LIFE CYCLE MANAGEMENT OF COMPLEX TECHNICAL OBJECTS
- ESTIMATION OF OPERATING COSTS WHEN CALCULATING THE TOTAL COST OF OWNERSHIP OF DISTRIBUTED TECHNICAL COMPLEXES
- SYSTEM ANALYSIS OF CLASSIFICATION AND ORDERING MODELS OF OBJECTS
- PLANNING THE RECONFIGURATION OF MULTI-MODE COMPLEX OBJECTS
- FUZZY-POSSIBILISTIC APPROACH TO MANAGING THE COMPLEXITY OF INTEGRATED INFORMATION AND CONTROL SYSTEMS
- MODEL AND ALGORITHMIC SUPPORT FOR FORECASTING AND PLANNING THE FORAGE HARVESTING PROCESS
- FORMATION OF THE CORE OF DOCUMENTS IN INTERNET MONITORING SYSTEMS UNDER RESOURCE CONSTRAINTS
- MODEL OF COMBINED APPLICATION OF INTELLIGENT METHODS OF INFORMATION SECURITY EVENTS CORRELATION