SBIR-STTR Award

TeraOps Corporation proposes to develop a suite of algorithmic techniques and corresponding hardware implementation based on the powerful CNN technology to address Navy requirements for multi-band image fusion in real-time. TeraOps will exploit the unique architecture of the CNN supercomputer-on-a-chip to implement spatial-temporal image processing algorithms for the fusion of multi-band sensory data to be used in advanced targeting systems. These algorithms will draw upon studies of biological image processing systems as well as on CNN models for color constancy and color visualization developed earlier by TearOps researchers. CNN is a massively parallel locally interconnected dynamic array computer capable of performing more than 10 to the 12 operations per second which allows it to perform image processing operations three orders of magnitude faster than conventional digital signal processors. Its very high processing speed is essential in executing computationally intensive image processing operations in real-time at high resolutions. The CNN image processing package including associated hardware will be small enough to fit in the limited space of an image processing pod and will consume very little power due to its highly efficient VLSI implementation. Thus, the proposed work combines the efficiency and speed of the hardware implementation and the algorithmic power of CNN technology to solve the real-time image fusion problem.

TeraOps will design and develop a working prototype of image fusion hardware and software capable of fusing up to 30 separate real-time video images. This ultra-high speed fusion system utilizes the outstanding capabilities of the Cellular Neural Network (CNN), a massively parallel analog array computer capable of operating at more than 1012 operations per second. The processor is extremely compact, about the size of a VCR cartridge, and will consume less than 10 W of power. The fusion algorithms will generate both grey-scale and false-color images of high quality and readability. The system will evaluate each input image for its information content at each point in space and time. The output image will maximize the information content across all input images. A universal coloring scheme will generate intuitively-natural colors for sky, land, foliage and other scene components.