Comparing Raster and Object Generalization Nigel Daley 1,2 , David G. Goodenough 1,2 , A.S. (Pal) Bhogal 1 , Quetzalcoatl Bradley 2 , and Z. Yin 2 1 Pacific Forestry Centre, 506 West Burnside Rd, Victoria, BC, Canada, V8Z 1M5 2 University of Victoria, Dept. of Computer Science, P.O. Box 1700, Victoria, BC, Canada, V8W 2Y2 Contact: ndaley@pfc.forestry.ca, 250-363-0153, fax: 250-363-0775 Abstract – Digital interpretation of imagery produces descriptions of the earth’s surface, each description relying on the inherent resolution of the original image. Forest cover geographic information (GIS) files have been produced by interpretation of aerial photography. Common mapping scales in Canada for representing land information are 1:20,000 and 1:250,000. This paper discusses two methods to automatically generalize GIS from higher spatial resolution scales to lower scales. These two methods are a raster method (MapGen) for generalization developed by Pamap and the BC Ministry of Forests, and an object-oriented method (ObjectGen). The GIS data set consists of topographic data and forest cover files, both at 1:20,000 scale and placed on the same datum. In this presentation we compare the results for generalizing forest objects by these different methods. This work leads to segmentations of remote sensing images, at corresponding resolutions to the GIS files, being used to constrain the generalizations. INTRODUCTION Remote sensing data is available from satellites and aircraft at multiple resolutions from 1 m to 1 km. For each application, there is a need to assess the utility of acquiring data at a wide variety of resolutions. In particular, can imagery at 1 m be used to derive digital interpretations appropriate for coarser resolutions? We chose to begin our investigation of how to generalize image objects by investigating the generalization of geographic information (GIS) files produced from interpretation of aerial photography. Our interest in this paper is in methods to automatically represent GIS information at other scales. We have implemented a raster method (MapGen) for generalization developed by Pamap and the BC Ministry of Forests (BCMOF) [1], and an object-oriented method described by Richardson [2]. The GIS data set consists of topographic data and forest cover files, both at 1:20,000 scale. In this presentation we compare the results for generalizing forest objects by these two different methods. The primary scales used for this test are 1:20,000 and 1:250,000. For remote sensing data, we will be segmenting images from the following sensors: MEIS (1 m), AVIRIS (20 m), TM (30 m), and AVHRR (1 km) for two test sites on Vancouver Island. THE DATA SETS Forest cover and hydrology GIS data are available for British Columbia at 1:20,000 scale. We worked with 3 mapsheets, 082E062, 082E072, and 083E073, which cover the Okanagan Mountain Park area just south of Kelowna, BC. Forest cover data came from BCMOF in IGDS/Forest Inventory Planning Data Exchange Format (FIPDEF). These data files were digitized from 1994 airphotos on the NAD27 datum. The Projected Type ID attribute is used by BCMOF to generalize forest cover data. Values and descriptions for this attribute can be found in table 1. Hydrology data (lakes and rivers) came from the BC Ministry of Environment, Lands, and Parks (BCMELP) Terrain Resource Information Management (TRIM) initiative. These data were digitized from airphotos using the NAD83 datum. They were converted to NAD27 to be compatible with the forest cover data. GENERALIZATION METHODS Two methods of automated generalization systems were compared: raster generalization (MapGen) and object generalization (ObjectGen). Raster Generalization MapGen, developed by Pamap (PCI Pacific) and BCMOF, is an automated raster generalization system. It is based on a set of polygon and vector generalization rules. Each polygon rule specifies how to combine neighbouring polygons. From Table 1. Generalized Classes & MapGen Rules Class Rule Min Size Merge List 0 Water 1 Immature (stocking class 0) PROJTYPEID = 1 15 3,2,4,9,5,6,8 2 Mature PROJTYPEID = 2 15 1,3,4,9,5,6,8 3 Immature Residual PROJTYPEID = 3 15 1,2,4,9,5,6,8 4 NSR PROJTYPEID = 4 15 9,1,3,2,5,6,8 5 Non Commercial PROJTYPEID = 5 15 6,4,9,1,3,2,8 6 Non Productive PROJTYPEID = 6 15 5,4,9,1,3,2,8 8 No Typing PROJTYPEID = 8 15 6,5,4,9,1,3,2 9 Silviculture NSR PROJTYPEID = 9 15 4,1,3,2,5,6,8