Computing the Visible Invariance in Grey Scale Imagery on the Transputer

In this paper we present a robust method for computing the visible invariance characteristics of 2D digitised grey-scale images. In essence visible invariance characterises features In the image plane, which are invariant to a specified set of transformations. The problem is formulated in variational terms and a least-squares technique is used to extract an optimum set of local features from the raw raster array of pixels. Appropriate mask operators for different window sizes to compute the image derivatives are given. These derivatives are then used for the computation of a new set of parameters, provided by classical differential geometry, the mean and Gaussian curvatures , which possess many invariant properties, and may be used to quantify the visible invariance. The method is parallel in nature and has successfully been implemented on the INMOS IMS T414 transputer running under an Occam environment. An analysis of the parallel implementation based on convolution type operator is also presented.