Determining the efficacy of edge detection algorithms

Edge detection is an important process in the interpretation of image data. Four types of edge detection algorithms (the Local Variance, Correlation, Laplacian and Frequency Peak algorithms) are compared for their ability to discriminate the edges present in an image with precisely defined edges. The variety of techniques are discussed in terms of their sensitivity to the presence of edges and their sensitivity to the applied noise. Both computational and empirical analyses are presented, and the relative merits of each discussed in terms of signal-to-noise ratio, sensitivity and computational difficulty. It is shown that local variance edge detection is an acceptable method for edge detection which simplifies the computation needed for image convolution. The means through which to devise more sophisticated edge detection algorithms is outlined, and the use of difference images to discriminate edges in an image is discussed. The results are also discussed in terms of the implementation of edge detection algorithms as part of a computer visual system which has an architecture modeled after the mammalian visual system.