Appearance-based target recognition and classification in infrared imagery

Our objective has been to find a preferred method for the identification of static targets in single IR images, concentrating on appearance-based methods. This has included thermal modelling of IR signatures and the identification of images of different objects with variation in pose and thermal state. Using principal component analysis, the variances among the images are extracted and represented in a low-dimensional feature eigenspace. Any new image can be projected into the eigenspace by taking an inner product with the basis. The object of interest can be recognized by a nearest-neighbour classification rule, made more accurate by application of over-sampling to the surface manifold by B-spline surface fitting, and made more efficient by a k-d tree search algorithm. To address the problems of recognizing targets in noisy and cluttered images, we have employed a random sampling approach that is based on the principle of high-breakdown point estimation. We have generated a database of images using visible and thermal cameras, in addition to scene simulation software, for use in the learning and recognition/evaluation phases. Our experiments indicate that application of the robust algorithm can reduce the recovery error of the true model image data, for example by a factor of five when the images contain 40% randomly changed image pixels.