Simulating infra-red land clutter: an image processing data source

In most image processing tasks, the essential difficulty is in the analysis- there is always a ready source of data. We describe a task where this is not the case, and our approach to resolving the difficulty. MIST has been designed to produce data to test image processing algorithms. Our system essentially approaches the image processing task back to front. We have attempted to develop a system which produces sufficiently realistic scenes to allow initial testing of algorithms without expensive data gathering trials. While not absolutely accurate, the system is useful for fast prototyping and testing ideas on algorithm modifications. The system models terrain using dynamic resolution techniques to reduce memory requirements, and models the components of the scene using a stochastic model. This model can generate an essentially infinite number of realisations of plausible scenarios, while taking into account heuristic rules on the arrangement of elements so that the scenes are realistic. The system has applications in mission planning, design and test of infra-red object detection algorithms, and in algorithm proving. Since the sequences generated by MIST can be exactly reproduced, they can also be used to provide a standard test suite for algorithm analysis.