Occluded Thermal Face Recognition Using Bag of CNN ($Bo$CNN)
Sumit Kumar, Satish Kumar Singh
Abstract
In this letter, we are proposing BoCNN architecture framework for occluded thermal face recognition. We analyzed the performance of Pretrained models using transfer learning and they exhibit promising results for thermal faces without occlusion. The performance degrades with occlusion. We have used different decision level fusion strategies post transfer learning for the performance enhancement of the pretrained models. All the fusion strategies used in the proposed work give better results compared to any of the single CNN architecture.
Topics & Concepts
Computer scienceTransfer of learningArtificial intelligenceFacial recognition systemFace (sociological concept)Pattern recognition (psychology)FusionOcclusionComputer visionConvolutional neural networkArchitectureVisual artsMedicineLinguisticsArtPhilosophySocial scienceCardiologySociologyFace recognition and analysisFace and Expression RecognitionVideo Surveillance and Tracking Methods