期刊名称：International Journal of Multimedia and Ubiquitous Engineering
印刷版ISSN：1975-0080
出版年度：2016
卷号：11
期号：12
页码：265
出版社：SERSC
摘要：Techniques of generating a single panoramic image by using multiple images are being widely studied in a number of areas, such as computer vision and computer graphics. Generating a panoramic image is a good way of overcoming the limitations of the images obtained from one single camera (e.g., those of picture angles, resolutions, information amounts, etc.) and may be applied in a variety of fields such as virtual reality and robot vision that require wide-angle images. A panoramic image has a great significance in that it can provide a greater sensation of immersion compared to a single image. Currently, there are a variety of techniques of producing panoramic images, but most of them commonly use a method of detecting feature points and matching points in each of the panoramic images they generate. In addition, they use the method of converting images after obtaining homography matrix using the RANSAC (Random Sample Consensus) algorithm that uses matching points. The SURF (Speeded Up Robust Features) algorithm used in this study utilizes the black-and-white and local space information of images when detecting their feature points and is widely used because it provides an outstanding performance in detecting the viewpoints and the changes of the image sizes and is faster than SIFT (Scale Invariant Features Transform) algorithm. However, the SURF algorithm also has its weak point of detecting wrong matching points, which may slow down the performance speed of the RANSAC algorithm and thus increases CPU usage occupation rates. The errors in detecting matching points serve as essential elements of lowering the accuracy and resolutions of panoramic images. In order to minimize these errors, this paper went through an intermediate filtering process of removing wrong matching points using the RGB values of 3×3 region around their coordinates and then presented analysis and evaluation results related to improvements in panoramic image construction & processing and CPU usage occupation rates and the decreasing rates and accuracy of the extracted matching points
