Feature based analysis of thermal images for emotion recognition

Thermal imaging has recently been investigated in automatic emotion identification to get an insight into reliable information about human emotion. However, the methods employed to classify thermal emotion in literature are discrete and randomly selected. These methods are deficit in explanation and also lack adequate justification for the obtained result. The assertions above are supported by the fact that, despite previous research, the existing methods are not successful in real-time and are resistant to obstacles such as spectacles, facial hair, and body movements. So there is enough room for more methods and thorough research into the effects of various features on thermal images and the characteristics of thermal emotion that are conveyed by those features. To address the issue, this research provides an in-depth performance analysis of hand-crafted features on thermal images while distinguishing emotion. In this study, we examine the inherent spatial and spectral aspects of several histogram-based feature descriptors along with a set of classifiers to classify thermal emotion. The study is carried out on two datasets, each with a distinct pseudo-color palette and sample size. Moreover, to the best of our knowledge, no work has been done to evaluate their feature extraction methods for the subject-independent case of thermal emotion recognition. Constructing a subject-independent model is the first step in classifying thermal emotive faces in real-time. This paper offers an early draft of the same, employing hand-crafted features. The author attempts to highlight the enormous scope of this research area because the subject-independent result obtained is weak. The existence of mixed emotions and inter-person variability are just two of the most likely causes of low accuracy and precision.