Estimation of optimal image object size for the segmentation of forest stands with multispectral IKONOS imagery
The determination of segments that represents an optimal image object size is very challenging in object-based image analysis (OBIA). This research employs local variance and spatial autocorrelation to estimate the optimal size of image objects for segmen
- PDF / 1,440,987 Bytes
- 17 Pages / 439.37 x 666.142 pts Page_size
- 51 Downloads / 223 Views
M. Kim1, M. Madden2, T. Warner3 1
Center for Remote Sensing and Mapping Science (CRMS), Department of Geography, University of Georgia, Athens, GA, U.S.A., [email protected]
2
CRMS, Department of Geography, University of Georgia, Athens, GA, U.S.A., [email protected]
3
Department of Geology and Geography, West Virginia University, Morgantown, WV, U.S.A., [email protected]
KEYWORDS: Image segmentation, very high spatial resolution image, over- and under-segmentation, object-based image analysis (OBIA), forest stands, local variance, spatial autocorrelation ABSTRACT: The determination of segments that represents an optimal image object size is very challenging in object-based image analysis (OBIA). This research employs local variance and spatial autocorrelation to estimate the optimal size of image objects for segmenting forest stands. Segmented images are visually compared to a manually interpreted forest stand database to examine the quality of forest stand segmentation in terms of the average size and number of image objects. Average local variances are then graphed against segmentation scale in an attempt to determine the appropriate scale for optimally derived segments. In addition, an analysis of spatial autocorrelation is performed to investigate how between-object
292
M. Kim, M. Madden, T. Warner
correlation changes with segmentation scale in terms of over-, optimal, and under-segmentation.
1 Introduction Conventional pixel-based classification approaches have limitations that should be considered when applied to very high spatial resolution (VHR) imagery (Fisher 1997; Townshend et al. 2000; Ehlers et al. 2003; Brandtberg and Warner 2006). The increased within-class spectral variation of VHR images decreases classification accuracy when used with the traditional pixel-based approaches (Shiewe et al. 2001). Object-based image analysis (OBIA), which became an area of increasing research interest in the late 1990s, is a contextual segmentation and classification approach that may offer an effective method for overcoming some of the limitations inherent to traditional pixel-based classification of VHR images. Particularly, the OBIA can overcome within-class spectral variation inherent to VHR imagery (Yu et al. 2006). In addition, it can be used to emulate a human interpreter’s ability in image interpretation (Blaschke and Strobl 2001; Blaschke 2003; Benz et al. 2004; Meinel and Neubert 2004). Although the OBIA scheme seems to hold promise for solving classification problems associated with VHR imagery, it also has an important related challenge, namely, the estimation of the desired size of image objects that should be obtained in an image segmentation procedure. Unfortunately, there is currently no objective method for deciding the optimal scale of segmentation, so the segmentation process is often highly dependent on trial-and-error methods (Meinel and Neubert 2004). Yet, the size of image objects is one of the most important and critical issues which directly influences the quality of the segment
Data Loading...
