Image Based Synthesis for Airborne Minefield Data

摘要

Image-based synthesis is a process of generating new images from already existing images such that newly created images are indistinguishable when compared with the real-world images. Minefield synthesis, in this framework, refers to the process of generating different realistic minefield scenarios from previously collected images of the mine, clutter, and background. In this thesis, minefield synthesis using patch-based sampling of previously acquired airborne MWIR (mid wave infrared) imagery is explored. The main idea is to synthesize a new minefield by selecting appropriate small patches from existing images and stitching them together in a consistent manner to simulate realistic imagery data representing different minefield scenarios. The selected patches include those from different background types in conjunction with natural and man-made clutter and different mine types. The background imagery is modeled using a first order Markov process, assuming that the statistical characteristics of image-patch at particular location are dependent only on the characteristics of patches in the adjacent neighborhood. A patch-based Markov model for different backgrounds is estimated from the available images. The proposed model is capable of generating any desired terrain condition (homogenous and inhomogeneous) from available background/terrain data. In addition, different minefield layouts such as patterned and scattered can be generated using mine patches. Minefield imagery for single frame as well as those simulating a flight path over a terrain map can be synthesized. Minefield image data synthesized using this procedure should be valuable for airborne mine and minefield detection program evaluate most mine detection as well as minefield detection algorithms.