Edge detection is required in a number of important image processing applications and the goal of edge detection is the detection and characterization of significant intensity changes. Among previously proposed edge detection operators, only Laplacian of Gaussian(LoG) edge detection operator which localizes edge through zero crossing in the filtered image is rotationally invariant. But some problems have not been solved. If standard deviation of LoG is small, then its edge location is almost exact, but unwanted edge is detected. Conversely, if standard deviation is large, then unwanted edge is not detected but its location is moved from real edge location. Another problem is spurious edge from nonlinear illumination. In this thesis, an iterative filtering by Laplacian of Gaussian is used to reduce to noise-like edges. Its performance is compared with Marr-Hildreth operator.
