Finding zero-crossings using the Laplacian of a LLMMSE filter for stationary random noise

Edge detection in the presence of noise is a fundamental problem in computer vision. Its solution directly influences the solution of many higher-level vision problems. A method for edge detection in the presence of stationary, signal independent, randomly distributed additive noise is presented. A rotationally symmetric Laplacian of a locally linear minimum-mean-squared-error (LLMMSE) filter is used to obtain the zero-crossings in an image instead of the Marr-Hildreth operator. The effect of this filter is studied for images degraded by Gaussian white noise.<>