STUDY OF EDGE DETECTION TECHNIQUES FOR TWO DIMENSIONAL IMAGES USING MATLAB

Abstract

Detection of edges is one of the most fundamental problems in the image analysis and computer vision. This dissertation work presents a comparative study of different techniques used for edge detection of gray-scale images. The edges are the regions in an image where the intensity value changes abruptly. Detection of edges is inevitable for obtaining the object boundaries in image segmentation. The detection of such boundaries is quite inherent in human visual perception. The edge detection methods based upon the first order derivative of image function are Sobel, Prewitt and Roberts. The application of these operators on sample test images are taken into account and their results are analyzed accordingly. The edges are detected as the regions of the local maxima of first order derivative of image function, and orthogonal to intensity gradient direction. The results of these operators on test images characterize computational simplicity and less execution time. Then Canny operator is applied on same sample images and the results are compared with the aforesaid operators. Further, the study of Lapacian of Gaussian (LOG) operator is carried out using the same test images and effects of variation of its parameters are examined. This operator produces the optimal trade off between two conflicting requirements in edge detection i.e. cancellation of noise and minimization of spatial smoothening. The study is further extended by introducing the Gaussian and Salt & Pepper noise in same test images respectively and the results of application of same operators are compared. Results so obtained establish the superiority of Canny and LOG operators over the classical operators in obtaining edges in noisy images. The processing of all test images for comparative study has been done by using the Image Processing Toolbox of MATLAB version 6.5.