Automatic Peak Recognition for Mountain Images
In this paper, we propose a novel method for automatically recognize the peaks of mountains based on the shape of skyline. Since the appearances of a mountain are variable due to the changes of weather, season or region, the skyline of mountains is extrac
- PDF / 1,846,401 Bytes
- 7 Pages / 439.37 x 666.142 pts Page_size
- 95 Downloads / 220 Views
Abstract In this paper, we propose a novel method for automatically recognize the peaks of mountains based on the shape of skyline. Since the appearances of a mountain are variable due to the changes of weather, season or region, the skyline of mountains is extracted for the matching of different mountain images. We use support vector machine (SVM) to predict the possible skyline segments and the linking of incomplete fragments of skyline is formulated as a shortest path problem and solved by dynamic programming strategy. In order to resist the geometric distortion caused by view change, we perform 2D curve matching on the extracted skylines for the peak recognition task. Our experimental results demonstrate that the proposed method is our method is effective for the mountain recognition under complicated and variable circumstances. Keywords Skyline localization
Mountain annotation
Introduction and Previous Work In last decade, many state-of-the-art image and visual subject retrieval approaches for place instance recognition [1, 2] and scene category recognition [3, 4] have been proposed. However, the automatic image-based location recognition for This work was supported by National Science Council, Taiwan, under the Grants NSC101-2221E-033-062. W.-H. Liu C.-W. Su (&) Department of Information and Computer Engineering, Chung Yuan Christian University, Chungli, Taiwan, Republic of China e-mail: [email protected] W.-H. Liu e-mail: [email protected]
Y.-M. Huang et al. (eds.), Advanced Technologies, Embedded and Multimedia 1115 for Human-centric Computing, Lecture Notes in Electrical Engineering 260, DOI: 10.1007/978-94-007-7262-5_127, Ó Springer Science+Business Media Dordrecht 2014
1116
W.-H. Liu and C.-W. Su
natural scene is still a challenging task. For example, the color and texture of vegetation may be very different due to the changes of seasons, weather and sunlight. It makes appearance-based techniques [1, 2] difficult to find the robust feature descriptors of a natural scene under different circumstances. In this paper, we consider the problem of recognizing the mountains in the image without geotagging information. Even for the geotagged photograph, only the position of the photographer is associated with the image in the most cases. To precisely annotate the names and positions of peaks in image, we extract skyline to locate the positions of mountains and then perform partial matching process on skylines to find the names of peaks. Currently, there is a limited number of studies focus on the image-based mountain recognition in recent years. In [5], Babound et al. detected mountains silhouette by compass edge detector and some post-processing steps. They compared the silhouette with a 3D terrain model of the mountains to extract silhouette accurately. The location of each mountain peak can be assigned accurately on images. In [6], Baatz et al. proposed a system for large scale location recognition based on digital elevation models. They extracted the sky and represent the visible horizon by a
Data Loading...
