Contents I. Inrodution
Edges in an image can be defined as the most important component as it forms the boundaries or outlines of the Region of Interest (ROI)s present in the image. It is the most commonly adopted approach in image segmentation which is the most basic step in pattern recognition research. By image segmentation, different objects in an image are isolated so that it catalyzes the further image analysis process. Edge detection is discontinuity based image segmentation approach [1]. Edge can be defined as a meaningful discontinuity of the image function [2]. So, the efficiency of any edge detection technique depends on how accurately these discontinuities are identified and brought out. In a gray level, an abrupt change of gray level leads to the formation of an edge. A gradient can describe this change in an image function [2] and hence it is required to measure gradient vector magnitude at pixel locations to detect an edge. The strategy of any edge detection technique can adopt either of the following two techniques: 1. Gradient-Based and 2. Laplacian-Based. The first one detects edges by finding maximum and minimum of the first derivative of the image, while the second one involves second order derivative of the image and the zero-crossings in the same will describe the potential edge locations[1] [3]. The second one involves high complexity than the first one. Also, in literature, it is found that maximum edge detections are done in gray images only. While a color image can bring much more important and hidden information than the same in gray form. Consequently, edge in a color image will be able to carry more prominent feature and thereby outlining a region of interest in a much attractive way than the same in the gray form of the image. This is the reason why we go for color image edge detection. Now, there exist a number of different definitions of color edge. What we follow here is the one given by G.S. Robinson in [4]. According to Robinson, an edge exists precisely in the color image if the intensity image contains an edge. Now, when we deal with a color image, the first important requirement that arises is the color space. Color space can be defined as an abstract model where exactly the color arrangements of the image are done [3]. RGB acts as a default color space in many applications. But, it will not be suitable for color image edge detection if we follow the definition of Robinson for the same as in RGB there is no any dedicated channel for intensity as each of the R, G and B channel carry color information as well as brightness values also. So, our main goal will be the selection of a color space where there is a specific channel for intensity information. HSV is such a color space where we have the V channel particularly devoted for the measurement of luminance level [3]. So, we have selected HSV color space for our color edge detection task. Now, as gradient-based techniques involve less complexity than Laplacian one, so we give preference the first one and selected Canny edge detector [5] for our approach. But, traditional Canny edge detector has some disadvantages, among them two major ones are: 1. Here it is needed to provide the size of the Gaussian filter manually whose value affects the results of the filter in a great way. There is no any adaptive technique to select this value automatically. Also, Gaussian filter while going for smoothing noises, it may sometimes smooth edges (high-frequency feature) also and hence it results in missing the identification of weak edges.
