NSF Award
8860331
High resolution scanned images are difficult to work with for several reasons: they require large amounts of storage and bandwidth, they are prone to aliasing artifacts when scaled or rotated, and operators cannot work directly and naturally on the objects depicted in the scene. We propose decomposing an image into 'surface patches' where each patch is described by an algebraic intensity function and a mathematical contour description. These patches will tend to correspond with objects in the scene, and are a compact and resolution independent representation of the image. The proposed work will investigate the feasibility of developing and implementing an algorithm to perform raster to object (image surface patch) conversion using present day or near future hardware technology. This will be done by conducting studies to characterize the image intensity functions and contours in a sample set of pictures. Candidate algorithms will be examined and promising ones prototyped to determine their potential for high fidelity image reconstruction and image compression.
