الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 148 |  |  |
| | | from | 148 | | |
Abstract
Image stitching ”mosaicing” has been a very important, challenging topic in computer vision, computer graphics, multimedia, and image processing. Image stitching is concerned with gathering / collecting two or more input overlapping images of the same scene into a large segmented image with a large field of view. Image stitching can be used for many different applications in the recent years. The most traditional application is the construction of large satellite photographs from collections of images. Another important application of panoramic image stitching is stitching medical images that can be obtained from several X-ray images and other applications in clinical diagnosis. There are two main types of approaches that are used for creating image stitching: Direct approach and Features-based approach. Direct methods compare all the pixel intensities of the images with each other, whereas feature-based methods aim to determine a matching “relationship” between input images through unique features extracted from the processed images. The greatest advantages of feature based methods over direct methods are their speed, robustness, and the ability to create a panoramic image of a scene however the existence of camera motion. Image stitching has many challenges such as the large amount of images may lead to high processing time, since each image needs some processing. Another challenge is the use of handled camera that may lead to presence of parallax; we need a high quality algorithm for stitching. In this thesis, we mainly aim to increase panoramic quality, decrease the error rate between overlapping images and decrease the execution time as possible. Therefore, we first compared many different feature-based detectors. The experimental results show that SIFT technique is a more robust algorithm but it takes the highest computation time. MSER and FAST techniques have better performance, faster, and more accurate. We concluded that ORB is the highest quality and the fastest. In addition, wediscussed the implementation of some different methods of image blending to get an increased quality of the stitching process. These methods are Exposure Compensation, Alpha blending, Gradient domain blending, and Laplacian pyramid blending.