Local Operations: Edging

Abstract

This chapter describes the boundary extraction algorithms based on local operators in the spatial domain and filtering techniques in the frequency domain. In the early stages of the vision process, some intrinsic characteristics are extracted from the image that are relevant to identify the objects of the scene. These elementary characteristics, called low level, are used by the visual system to isolate the objects of the scene, perceived as relatively homogeneous regions with respect to one or more of the following attributes: gray level (intensity), color, texture, distance, motion, nature of the material, surface condition of objects, etc. In analogy to visual perception, for the artificial system, the acquired image of the scene presents itself groups of regions with abrupt discontinuities evaluated among themselves that can be assessed on the basis of some of the attributes indicated above. In this chapter, we are interested in the analysis and search of image pixels where these discontinuities occur, which are typically associated with the edges of the objects, that is, boundary zones between regions of the image with different attributes, or, more in general, for the discontinuity of reflectance and different illuminations of objects. All these causes make the edge extraction operation not easy, considering also the intrinsic noise of the image pixels. The operation of extracting the edges is still today a fundamental task for the analysis of the images. In this chapter, the most common algorithms will be described to determine the edges with local operators known in the literature as Local Edge Detector—LED. These algorithms assign a value to each pixel automatically evaluated as an edge element, but no information is generated to link the various edge pixels (link edge) together to form the edge segments. Other algorithms will be described in the following to link together the edge pixels which belong to the same contour. The LED algorithms are local operators that determine local variations through a direct analysis of the gray-level values of the image or through local variations of the derivatives of the intensity function. These local discontinuities near the edges can be of various types. A brief description of these discontinuities and a graphical representation on the edge profile is given.