II.11 - NOISE THRESHOLDING IN EDGE IMAGES

Abstract

This chapter analyzes noise thresholding in edge images. First derivative filters (often called edge detectors) are used to sharpen digitized images, or to produce a candidate set of points at which to evaluate some other operator. They are also used to obtain a surface normal perturbation from an image for bump mapping. Digitized images are typically noisy. First derivative estimators such as the Sobel and Prewitt operators accentuate this noise. Hence, it is often necessary to threshold the result of the edge operator to filter out the noise. Two most common first derivative edge detectors are the Sobel and Prewitt operators. The 3-by-3 convolution masks are evaluated at each pixel to produce estimates of the partial derivatives of the image function. The edge function E(p) produces an estimate of the magnitude of the gradient. The threshold can be based on an evaluation of the noise in the gradient domain. The chapter also discusses the case of bump mapping.