Fractal Image Sensor

Fractal mathematical concepts are useful for modeling natural image backgrounds (foliage, clouds, sand dunes) and can help discriminate target from background in IR/thermal images on the basis of fractal dimension - particularly useful in a multiband data environment requiring sensor fusion. Similar possibilities exist for data returned from millimeter wave radar guidance. To date, however, fractal characterization of captured image data has been considered too computationally intensive for real-time respone. Foster-Miller has develped a real-time "fractal image sensor" which instantaneously measures fractal dimension by using a parallel optical coprocessor based on photorefractive non-linear optics. By reducing complex image textures to a small fractal feature vector, electronic neural networks can also be applied to the coprocessor output for training on more subtle fractal characteristics. Phase I research will test the applicability of the fractal optical coprocessor to a database of typical IR/thermal imagery and also evaluate the potential for an all-electronic implementation. Phase II will construct a complete hybrid optical/electronic thermal image computer incorporating image segmentation and analysis, fractal modeling and synthesis, a dedicated fractal coprocessor for high speed characterization, and neural network training.