Extracting Facial Skin Information for Temperature Measurement Using RGB-Thermal Image

This paper presents an exploration of facial skin temperature responses while the subject’s head posture is simultaneously varied in indoor environment settings by using benchmark datasets that include thermal and RGB images from healthy groups. The selection of gender-balanced was considered in the experiment, and thus an image enhancement algorithm was proposed before the image quality assessment is performed to characterize the quality score for individual images. For automatically extracting facial skin temperature, two sets of the labeled region of measurements are consistently detected across both image spectrums for all subjects. The mean temperature is then calculated in each labeled region and found to be statistically different due to the correlation of color temperature produced by each image group. In this work, the utilization of large image datasets containing a range of temperature trends may considerably improve temperature estimation for early detection of health issues. Although the findings may not be as accurate as the use of biosensor methods to represent actual body temperature, the designated algorithms manage to improve the analysis success by showing warm or cold temperature values as a quick screening of human spontaneous that may be correlated to emotional discomfort for contactless use. The experiments demonstrate that a statistically significant change in the temperature measurements can be found between the selection of facial regional, gender-related and asymmetry analysis on similar facial anatomy. Recent studies indicate that these parameters continue to be a significant basis for instigating of the risks associated with cardiovascular disease (CVD) in clinical settings.