Texture Analysis of Blood Vessels in Thermal Image for Affective States Classification on Children

Abstract

The maturity of affective computing via thermal camera is evolving as the technology advances in detecting the radiating heat from the human face, indicating the area concentrated with blood vessels. The capability of the thermal camera as non-invasive tools for thermal data recording is well established and highlighted. However, the research on autistic children’s affective states classification by using thermal camera has yet to be investigated to our knowledge. The autistic children have difficulty in affective states presentation through facial expression. Hence, it is hypothesized that the cutaneous temperature changes in the blood vessels would have direct impact on the affective states. In this work, healthy children were recruited as subjects prior to developing the reference model algorithm in thermal imaging analysis. A Mean of Correlation (MoC) of the Gray-Level Co-occurrence Matrix (GLCM) method was applied in feature extraction stage to produce the classifier’s best prediction performance. The accuracy of the weighted k-Nearest Neighbor (k-NN) classifier was recorded at 88.0%. The proposed method suggests that these analyses are significant for distinguishing between five basic affective states, and its applicability to autistic children.