Parallel vision algorithm design and implementation : 1988 end of year report

Abstract : The Apply programming language has been extended to allow variable- sized image computations, and also to allow border mirroring, in which pixels accessed outside the borders are produced by copying pixels from the interior of the image. Implementation and design decisions are discussed. Apply and the Warp programming language W2 were used to implement the second DARPA image understanding benchmark. The results of this implementation are reported. Experience with this benchmark suggests a method for performing global image computations in a machine independent manner, using the divide and conquer model. Implications of this model for algorithms in the image understanding benchmark are discussed. It is shown that this model is capable of computing any algorithm in which data is accessed in a fixed order, regardless of the data values, and in which the final computation is reversible: that is, it produces the same results if the data values are reversed in order.