ABSTRACT
Edge detection is a fundamental low-level image processing. An edge is a property of an individual pixel, which is calculated from the functional behaviour of an image in its neighbouring pixel. Edge serves in simplifying the analysis of images by drastically reducing the data to be processed and preserving the useful structural information about object boundaries. Edges in an image are the areas with strong intensity contrasts – a jump in intensity from one pixel to the next pixel. There are many ways to perform edge detection, but the majority of methods may be grouped into two categories: (a) gradient and (b) Laplacian. Gradient-based methods detect the edges by looking at the minimum and maximum of the first derivative of the image. Laplacian method searches for the zero crossings in the second derivative of the image in order to find edges. A number of edge detection techniques may be found, but there is not a single method that can detect the edges efficiently. Traditional edge detection methods such as Robert, Sobel and Prewitt operator and Laplacian of Gaussian operator are widely used. Canny proposed an optimal operator for edge detection. Marr and Hildreth’s [17] method is based on the zero crossing of the Laplacian operator, which is applied to the Gaussian smoothed image. But in the detection of the zero crossings in the second derivative, the maxima of the gradient are also captured and give false edges. Most of the existing techniques either are very sensitive to noise or do not give satisfactory results in low-contrast areas.
