Please use this identifier to cite or link to this item: http://dspace.dtu.ac.in:8080/jspui/handle/repository/21247
Title: EDGE DETECTION IN DIGITAL IMAGES USING SOFT COMPUTING TECHNIQUES
Authors: RAHEJA, SAHIL
Keywords: EDGE DETECTION
DIGITAL IMAGES
SOFT COMPUTING TECHNIQUES
FUZZY LOGIC
DNN
ANN
Issue Date: Dec-2024
Series/Report no.: TD-7633;
Abstract: Analyzing the contents of a real-world view is an essential task which needs to be carried out in machine vision and image processing, that has gained a lot of interest from the researchers in the past four decades. Edge detection algorithm calculates the contours of an object, separates it from its background and helps in analyzing the contents of the image. This represents the significance of edge detection in the area of machine vision. The method of edge detection involves in locating the points in a digital image where abrupt changes in image intensity or discontinuities occurs. There are different conventional algorithms useful for edge detection in various points of view. Some of the popular algorithms are Sobel, Canny, and Roberts etc. All of these traditional methods are based on a gradient estimation for the pixels. These techniques are less complex but fail in the presence noise. These techniques are highly dependent on the threshold which results in missing, false and spurious edges. Several other techniques are also proposed in the literature. For edge detection, soft computing based techniques had recently gained much popularity. In these methods, artificial neural networks (ANN), Fuzzy logic, deep neural network (DNN) and evolutionary algorithms (EA) based on swarm’s behaviors’ are the most common. The ANN, and DNN based methods are complex and very susceptible to noise perturbations. In some of soft-computing methods, weak edges are also not detected properly. To overcome these limitations, a two step process for edge detection is considered. In the first step, edge refinement is done, and then edge detection is done using soft computing technique. In our first work, an edge detection method based on Ant Colony Optimization (ACO) is described. A novel intensity mapping function is utilized to record intensity change among neighbouring pixels which guides the movement of ant. Finally, the Peak Signal-to-Noise Ratio (PSNR), accuracy, and F-Score are used to evaluate and compare performance. In the second work, fuzzy logic-based edge detection approach using a sharpening guided filter is proposed. A Gaussian filter is also used to deal with noise caused by sharpening. A range of statistical indicators are used to assess the method's accuracy. ix It has been discovered that by properly setting the smoothening parameters, a significant improvement can be achieved in the identified edges. Simulation results are presented on various images, statistical results are evaluated and compared with other latest techniques. In the third work, images are smoothed at varying degrees using a guided L0 smoothen filter and then a fuzzy logic-based edge detection algorithm is used to detect edges. Simulation results for Canny, Sobel, fuzzy logic-based edge detection, and lastly fuzzy logic edge detection with L0smoothen filter are shown. Results are contrasted with both traditional and contemporary methods. More than 100 images are taken into account from the Berkley Segmentation Database (BSD) and USC-SIPI Image Database. The measured F-value reaches a maximum of 0.848. Finally in fourth work, the edge detection considering guided image filtering and ACO is discussed. Simulation results are presented on various images and statistical results will also be evaluated and comparisons with latest techniques are also done. In this research, an effort is made to propose an edge detection algorithm which can work well in presence of noise. The performance of proposed edge detection techniques is analyzed with other notable techniques. The simulation is performed on BSD (Berkeley Segmentation Dataset) and USC-SIPI Image Database using computer simulation in MATLAB. Summarized table for result using different parameters achieved by proposed method and other edge detection techniques shows that the proposed method improves the performance of edge detection. Due to better performance of proposed method; it can be used in different field of science and engineering. In addition, possible directions for further developments are outlined.
URI: http://dspace.dtu.ac.in:8080/jspui/handle/repository/21247
Appears in Collections:Ph.D. Information Technology

Files in This Item:
File Description SizeFormat 
Sahil Raheja Ph.D..pdf6.61 MBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.