Reduced isothermal feature set for long wave infrared (LWIR) face recognition

• Published in: Infrared physics & technology Vol. 83; pp. 114 - 123
• Main Authors: Donoso, Ramiro, San Martín, Cesar, Hermosilla, Gabriel
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.06.2017
• Subjects: Expectation Maximization (EM); Face recognition; Isothermal imaging; Face recognition; Expectation Maximization (EM); Isothermal imaging
• ISSN: 1350-4495; 1879-0275
• DOI: 10.1016/j.infrared.2017.04.019

•A new methodology for long wave infrared face recognition is proposed.•An extraction methodology of reduced isothermal features set is proposed.•A probabilistic proximity index for comparison between features set is proposed. In this paper, we introduce a new concept in the thermal face recognition area: isothermal features. This consists of a feature vector built from a thermal signature that depends on the emission of the skin of the person and its temperature. A thermal signature is the appearance of the face to infrared sensors and is unique to each person. The infrared face is decomposed into isothermal regions that present the thermal features of the face. Each isothermal region is modeled as circles within a center representing the pixel of the image, and the feature vector is composed of a maximum radius of the circles at the isothermal region. This feature vector corresponds to the thermal signature of a person. The face recognition process is built using a modification of the Expectation Maximization (EM) algorithm in conjunction with a proposed probabilistic index to the classification process. Results obtained using an infrared database are compared with typical state-of-the-art techniques showing better performance, especially in uncontrolled acquisition conditions scenarios.