On Human Faces Thermal Image Processing for Classification Purposes

This paper presents some results from a preliminary work on human faces thermal image processing for classification purposes. The main task is to find right representations and descriptors/features for high recognition rates. A data base with thermal images is created using people in various states, which reveals happiness, normal and sadness states. A toolbox based on Matlab modeling and simulation environment was designed and developed. The toolbox has specialized functions for image conversion from color to gray levels, format image conversion, image segmentation, objects detection and localization, face contours, image compression and analysis. Various data transform for image processing are considered, accompanied by information-based transforms, e.g. Renyi transforms. Classification is considered and it is based on similarity of the patterns. The results are very encouraging and show the flexibility of the toolbox and the potential of the proposed methods based on Discrete Cosines Transform (DCT) for feature selection and extraction.