EDGE DETECTION USING EVOLUTIONARY ALGORITHMS

Abstract

The problem of edge detection in digital images is addressed using two methods. The first method proposed for edge detection in color images is based on fuzzy logic, and evolutionary learning techniques such as bacterial foraging, particle swarm optimization, genetic algorithm and gravitational search algorithms. In this method a circular mask is applied separately on R, G, and B components of RGB image to compute USAN area about every pixel in the image leading to the USAN area image. Histogram based Gaussian membership function is used to detect strong edges and the modified bell shaped membership function is used to detect weak edges from this image after fuzzification. Fuzzy entropy and edge sharpness factor involved in the objective function and the parameters that control the shape and range of modified bell shaped membership function are optimized using different evolutionary techniques. The edge maps of the three color components are obtained by subjecting the fuzzified USAN area images of color components to adaptive threshold. The final edge map is obtained by combining the edge maps of the above three color components. The second method proposed for grayscale images is based on fuzzy derivative and evolutionary algorithms. The derivatives in all eight directions for edge pixel are computed using if-then rules. Median of the derivative set is taken to obtain derivative matrix which is fuzzified using bellshaped membership function. Another membership function HIGH acting as hedge is used to locate the edges in the image in the fuzzy domain. Fuzzy entropy involved in the objective function and parameters that control the shape and range of bell shaped and HIGH membership functions are optimized using different evolutionary algorithms. Adaptive threshold is applied to fuzzified image to obtain the final edge map. Visual comparison is performed between the experimental results of the two methods to understand their efficacy.