Thermophysical algebraic invariance for infrared image interpretation
N. Nadhakumar, Jonathan Michel · 1996
An important issue in object recognition is the specification of features that are: (a) invariant to viewing and scene conditions, and also (b) specific, i.e., the feature must have different values for difference classes of objects. This dissertation presents a new approach for computing invariant features from infrared imagery. The approach is unique in the field since it considers not just surface reflection and surface geometry in the specification of invariant features, but it also takes into account internal object composition and state which affect images sensed in the non-visible spectrum. A new type of invariance called Thermophysical Algebraic Invariance (TAI) is defined. Features are defined such that they are functions of only the thermophysical properties of the imaged objects. The approach utilizes a physics-based model that is derived from the principle of the conservation of energy (COE) applied at the surface of the imaged object. Investigation into the specification of invariant features from different polynomial forms of the COE equation yield three basic invariant forms. The first, from algebraic invariance theory, is defined as a ratio of relative invariants. The second and third, using a more general approach, are derived using symbolic algebraic elimination applied to linear and non-linear forms of the COE equations, respectively. The features identified by these methods are independent of scene-to-scene transformation of the driving conditions such as ambient temperature, and wind speed and solar insolation. A context-based approach to using the TAI features defined above yields a method where features are constructed with as few as two points from the imaged object. Such a formulation can support hierarchical recognition schemes. The computation of TAI features requires correspondence to be established between a thermophysical object model and the imaged region. This dissertation describes the use of Geometric Invariants to generate hypotheses of object identity and pose to establish point correspondence. The hypotheses are verified or refuted by the thermophysical features. Thus, an integrated approach is developed for the model-based recognition of objects in IR images that combines TAI and geometric invariant features. Results on real IR imagery are shown illustrate the performance of the approach.