Аннотация:Since infrared images reflect the temperature-field distributing of vein¿s embranchment of the face, expression recognition from them should focus on changes of thermal distribution on facial muscles and blood vessels, which caused by mood variety. These thermal changes can be regarded as texture features of images. Wavelet transform is a good tool to analyze multi-scale, multi- direction changes of texture. Therefore, this paper presented a facial expression recognition method from thermal images using wavelet transform. Firstly, the status quo of the researches on infrared facial expression recognitions was briefly surveyed and analyzed. Secondly, wavelet transforms were used to extract image features. Thirdly, the ¿Euclidian Distance Maximizing Method¿ we proposed was adopted to select features. Then principal component analysis was employed to reduce feature dimension and NN was used as a classifier. Finally, experiments on medium-wave thermal facial expressions database of Equinox evaluated the effectiveness of our approach.
