Neural Geometric Engine Using Cortical Information Processing

CST proposes research and development of an artificial neural geometric engine designed according to a Lie group model of cortical visual information processing that can rapidly determine affine difference between two images. Computing parameters of geometric transform, such as scale, rotation, shear, shift, are considered as "inverse problem" in computer vision. The currently available warping and matching algorithm, which determine geometric distortion by trial and error, is very time consuming. The Lie group model of cortical visual information processing is a dynamical system. The neural geometric engine equipped with Lie germs is capable of computing Lie derivatives of imagery by linear filtering, and further determining affine differences of two images in a dynamical system. The full development of a Lie group model neural geometric engine will solve the geometric transform problem in realtime and will have very extensive applications in the area of imaging sensor data processing and will have great impact to artificial vision. Successful development of artificial neural geometric information processing system will make possible for quick terrain scene visualization from raw satellite and aerial photographs. The new technology can be applied to computer stereometry, image motion compensation and stabilization for automatic guided vehicles, and invariant pattern analysis for OCR.