Hardware Implementation of a Real-Time Image Segmentation Circuit based on Fuzzy Logic for Edge Detection Application

Abstract

Edge detection algorithms in images make it possible to extract information from the image and reduce the amount of required stored information. An edge is defined as a sharp change in luminosity intensity between two adjacent pixels. Most edge detection techniques can be grouped into two categories: gradient based techniques and Laplacian based methods. Techniques based on gradient use the first derivative of the image and look for the maximum and the minimum of this derivative. Examples of this type of strategies are: the Canny method (Canny, 1986), Sobel method, Roberts method (Roberts, 1965), Prewitt method (Prewitt, 1970), etc. On the other hand the techniques based on Laplacian look for the cross by zero of the second derivative of the image. An example of this type of techniques is the zero-crossing method (Marr & Hildreth, 1980). Normally edge extraction mechanisms are implemented by executing the corresponding software realisation on a processor. Nevertheless in applications that demand constrained response times (real time applications) the specific hardware implementation is required. The main drawback of implementing edge detection techniques in hardware is the high complexity of the existing algorithms. The process of edge detection in an image consists of a sequence of stages. Image segmentation is one step in the edge detection process. By means of the segmentation the image is divided in parts or objects that constitutes it. In the case of considering only one region the image is divided in object and background. The level at which this subdivision is made depends on the application. The segmentation will finish when all the objects of interest for the application have been detected. The image segmentation algorithms are based generally on two basic properties of the image grey levels: discontinuity and similarity. Inside the first category the techniques tries to divide the image by means of the sharp changes on the grey level. In the second category there are applied thresholds techniques, growth of regions, and division and fusion techniques. The simplest segmentation problem appears when the image is formed by only one object that has homogenous light intensity on a background with a different level of luminosity. In this case the image can be segmented in two regions using a technique based on a threshold parameter. Thresholding then becomes a simple but effective tool to separate objects from the background. Most of thresholding algorithms are initially meant for binary thresholding. This binary thresholding procedure may be extended to a multi-level one with the help of