You are on page 1of 1

NAME: DEMBE OSCAR REGNO: 18/U/19807/PS

Edge Detection and Segmentation of Images

The core function for image processing is the ability to identity the boundary between the objects in an
image and separate the objects from each other. Segmentation and edge detection in many applications
allow separation of objects in an image and identify the boundaries around them. The examples of edge
detection include sobel edge detection, laplacian of gaussian edge detection, and canny edge detection.
The examples of image segmentation techniques are point detection, line detection, and region and
object segmentation.

Sobel Edge Detection is a computationally simple technique that uses a 3 × 3 simple mask to magnify
the differences among the points on the opposite sides of a boundary and to eliminate the smooth gray-
level changes in the pixels located on the same side of the boundary. The Sobel mask can detect both
vertical lines and horizontal lines. A combination of the horizontal edge-detecting and vertical edge-
detecting masks must be used to capture the edges for this technique to be accurate. For small scale
applications, any edge can be approximated by a number of short horizontal and vertical edge
components that can be detected. In Sobel edge detection, the image is often divided into smaller
blocks, and then, in each block, a combination of vertical and horizontal Sobel edge detection is
performed. Then, the detected edges in smaller blocks are combined with each other to form the edges
in the complete image.

Laplacian of Gaussian Edge Detection is an edge detection technique that combines the Laplacian
operator and a Gaussian smoothing filter. The use of a Laplacian operator for edge detection is because
of weak edge detection since if not applied they cannot be detected. The first derivatives that are used
in masks such as Sobel may not be large enough to distinguish the edge points, hence a weak edge can
go undetected. Taking the second derivative (Laplacian) amplifies the changes in the first derivative
and therefore increases the chances of detecting weak edges. The Laplacian if used alone can magnify
noise in the image which can lead to detection false edges thus a Gaussian smoothing filter, is used to
filter out the high-frequency noise before applying Laplacian operator. In Canny Edge Detection, the
image is smoothed using a Gaussian filter, the gradient magnitude and orientation are computed using
finite difference approximations for the partial derivatives, non-maxima suppression is applied to the
gradient magnitude to search for pixels that can identify the existence of an edge, a double thresholding
algorithm is used to detect significant edges and link these edges.

Image segmentation can be done with two approaches where by one can do it by using the
dissimilarities among neighboring pixels and the other method would be to do it using the similarities
of neighboring pixels and thus the first approach uses methods like detecting gray-level discontinuities
with approaches like line, point and edges and thresholding and also Region and object segmentation
includes region segmentation using luminance thresholding. In the second category we have region
growing that uses a seed pixel and quad trees that don’t depend on seed pixel set which divides the
image into a set of disjointed regions and uses splitting and merging of pixels/ regions to obtain the
segmented regions that satisfy a given criterion.

You might also like