Infrared Thermography (IRT) has been used to monitor skin temperature variation in a contactless manner, in both clinical medicine and psychophysiology. Here, we introduce a new methodology to obtain information about autonomic correlates related to perspiration, peripheral vasomotility, and respiration from infrared recordings. Our approach involves a model-based decomposition of facial thermograms using Independent Component Analysis (ICA) and an ad-hoc preprocessing procedure. We tested our approach on 30 healthy volunteers whose psychophysiological state was stimulated as part of an experimental protocol. Within-subject ICA analysis identified three independent components demonstrating correlations with the reference physiological signals. Moreover, a linear combination of independent components effectively predicted each physiological signal, achieving median correlations of 0.9 for electrodermal activity, 0.8 for respiration, and 0.73 for photoplethysmography peaks envelope. In addition, we performed a cross-validated inter-subject analysis, which allows to predict physiological signals from facial thermograms of unseen subjects. Our findings validate the efficacy of features extracted from both original and thermal-derived signals for differentiating experimental conditions. This outcome emphasizes the sensitivity and promise of our approach, advocating for expanded investigations into thermal imaging within biomedical signal analysis. It underscores its potential for enhancing objective assessments of emotional states.