On Change Detection in the State of the Human Faces by Thermal Images and Renyi Information Measures

摘要

The objective of the work is to evaluate two measures of information content in some areas of human face thermal images, for change detection in the state, i.e. which express normal, sad, or happy states. The motivation of such investigation is given by the high interest in thermal image processing, as alternative to visual images, with direct applications in various fields of engineering, such as e.g. authentication, and diagnosis. The Renyi entropy of high order and Renyi divergence of order 0.5 are used. These information measures could respond to the requirements of low complexity of image processing and later to a small cost of the implementation circuits/system. For each measure, two approaches for model estimation of probability density function (pdf) are considered. The first one is based on the source identification (as symbols and probabilities) and, the second one, directly by image normalization. The preliminary results obtained with a small data base composed of 21 thermal images are satisfactory and indicate better results for the Renyi divergence and pdf estimation based on the normalization of the thermal images.