A machine process for identification and extraction of magnitude and direction of edges and lines in noisy multidimensional imagery data. A digital picture function is viewed as a sampling of the underlying reflectance function of the objects in the scene or pattern with noise added to the true function values. The edges or lines relate to those places in the image where there are jumps in the values of the reflectance function or its derivatives. By expressing the function in some parametric form in the local neighborhood of the pixel under consideration, the edges or lines and the types of edges (left to right or right to left) or lines (peak to trough) may be inferred from the values of the parameters. Assuming the noise is Gaussian, significant edges or lines are detected by performing the statistical hypothesis tests on the parameters of the function. Recursive relations are used for efficiently estimating the parameters of the function.
展开▼
