Model-based image enhancement of far infra-red images

In this paper we devise robust and reliable image enhancement algorithms for far infra-red images. The algorithms are based upon a model of the imaging process. The model that we develop is qualitative due to the unpredictable nature of the scenes. The model predicts that an idealized far infra-red image consists of piecewise-constant patches. On the basis of this model and models of the degrading factors we then apply two known algorithms: median filtering and spatial homomorphic filtering. In order to speed up the enhancement we extend our model temporally and develop a new algorithm: spatio-temporal homomorphic filtering. The paper ends with an attempt to convert the far infra-red images taken at night to visible daylight images