Segmentation of Infrared Images Using Cued Morphological Processing of Edge Maps

The segmentation of infrared images presents some unique challenges and depends greatly on the application considered. Methods developed for military FLIR or predictive maintenance applications do not perform well when translated to other applications such as infrared imaging of the human body. One solution to this problem is to use a small amount of available a priori information about the expected shape of the object to be recovered, in order to constrain the segmentation. In this paper, we present a new approach for segmenting infrared images in challenging environments, using a constrained morphological processing of edge maps. This method allows to recover weaker contours in noisy and non-uniform backgrounds. Results on synthetic and real objects show that the proposed approach outperforms other current infrared image segmentation algorithms for varying contrast and noise levels. Applications of our method include automated extraction and analysis of temperature variations in infrared image sequences with minimal or no interaction from a user.