Efficient emotion recognition from thermal faces using histogram and Hu features

The current trend in analysing bodily functions and understanding affective and psychological needs through a contactless approach is the use of thermal IR imagery in physiological and affective computing. Thermal imaging could upgrade the visibility of objects in a dark environment by identifying the objects & infrared (IR) radiation and making an image depending on that information. Facial tracking in the IR spectrum enhances interest in the current research scenario, especially in the applications like surveillance and face detection, because IR imaging is independent of lighting conditions. Changes in Facial thermal image are due to an emotional state in addition to facial expression changes because thermal imaging technology unveils the heat energy radiated from the face. This research work was initially based on facial thermal images from a standard database. A popular benchmark database is the iVITE Database. A database termed Malnad Thermal ECG and Respiration (MATHER) Database is developed for conducting experiments. Frontal face thermal images are used in all the experiments. Facial Landmark Points are used to develop regions corresponding to eyes, cheeks, nose and mouth from which histogram features are derived, an accuracy of 83.92% is obtained. An effort is made to extract facial image directly from thermal image using HSV space and colour thresholding after that features based on Hu’s Moments and Histogram Statistics are computed to yield the highest recognition rate of 87.33%.