AUTOMATED COUNTING SOFTWARE An image-processing program for automated counting David J. Cunningham, William H. Anderson, and R. Michael Anthony Address for David J. Cunningham (and for William H. Anderson during this research): Center for Mapping, The Ohio State University, 1214 Kinnear Road, Columbus, OH 43212, USA. Current address for William H. Anderson: Daedalus Enterprises, Inc., PO Box 1869, Ann Arbor, Ml 48106-1869, USA. Address for R. Michael Anthony: Alaska Science Center, National Biological Service, 101 1 East Tudor Road, Anchorage, AK 99503, USA. Key words: aerial survey, image processing, remote sensing, software Counts of dense concentrations of birds from aerial photographs have been employed widely (Kadlec and Drury 1968, Heyland 1972, Leonard and Fish 1974, Kerbes and Moore 1975, Harris and Lloyd 1977). More recently computers have been used to improve efficiency and precision of counts from digitized images of birds (Gilmer et al. 1988, Bajzak and Piatt 1990). To create a more interactive environment than these automated counting programs provided, we adapted Macintosh-based, public-domain software for counting black brant (Branta bernicla nigricans) in large flocks staging on water. This software accesses and enhances digitized images; produces tabular output of data; and allows interactive selection of counting parameters that include spectral-reflectance values, size, shape, and orientation. Efficient application of this program depends on high contrast between objects of interest and background, large concentrations of objects of interest, separation of individuals, and consistent image quality. Manual counts can be taken by marking objects with a paint tool; marks are counted automatically. The program does not have pattern recognition capabilities but had a mean (+ SE) counting error of 4.17+1.89 percent in a test count of 1,750 birds with 16 images of brant, emperor geese (Chen canagica), and snow geese (Chen caerulescens) when compared to manual counts from the images. Description of software Our automated counting program, DUCK HUNT, is an adaptation of IMAGE, a public-domain image-processing and analysis program developed by the National Institutes of Health (NIH, Bethesda, MD 20892). We have used it to analyze digital images acquired from scanned photographs, captured video images, photoCD's from 35 mm transparencies, and image files from digital cameras. It operates on Apple Macintosh computers Apple Computer, Inc., 1 Infinite Loop, Cupertino, CA 95014-2084) and reads and writes in TIFF, PICT, PICS, and MacPaint file formats. Objects of interest for counting e.g., geese on water) must have adequate contrast with their background and be spatially unique for the count to be accurate. Marginally defined images can be enhanced through spatial and spectral processing. These operations are conducted interactively through pull-down menus. Enhanced images can be processed immediately with counting routines or saved to disk. Following the approach of Gilmer et al. 1988), combinations of manual and automated counting functions also can be used to estimate mean pixels/bird. Objects to be counted are selected by their spectral reflectance, (in pixel groups numbered 1-254) using interactive density slicing of the image. When density slicing is enabled, the entire range of digital numbers is depicted in a rectangular look-up-table (LUT) window adjacent to the image window. Digital values are selected by using the mouse cursor to adjust a colored bar that overlaps the desired range within the LUT window. Pixels with digital values overlapped by the colored bar are highlighted with the same color in the image. As the density slice is manipulated, the image is dynamically updated to show which objects are highlighted. By changing the width and location of the colored bar, an appropriate range of digital values that highlights only the objects of interest can be found. When all objects of interest have been highlighted, the mouse cursor is used to mark a sample of highlighted objects that define selection parameters for objects of interest (e.g., geese). The sample should include objects of interest with sizes and shapes at the extreme limits of the image population (i.e., smallest, largest, thinnest, fattest). When a sample of objects has been selected, the counting routine labels each object with a consecutively numbered tag and constructs an exportable table with size and shape parameters for each object. To provide an adequate range of objects there is an option to count objects within 2 standard deviations of the mean size. Objects within this range are labeled with a black, numbered tag; those within the selected range have a white tag. The process of selection and counting can be repeated until the operator is satisfied that all objects of interest have been counted. The selection parameters used for counting can be saved and used for processing other images with similar optical characteristics. Hardware requirements IMAGE 1.45CFM requires an Apple Macintosh computer with at least 4MB of memory, but >=8MB are recommended for using DUCK HUNT to analyze large, high resolution images. It requires a monitor with the ability to display 256 colors or shades of gray. It reads and writes 8-bit TIFF, PICT, PICS, and MacPaint file formats. A Pascal-like macro programming language can be used to automate complex and repetitive tasks. Copies of the program, documentation, and source code can be obtained via anonymous FTP from zippy.nimh.nih.gov (ip address 128.231.98.32) or by contacting R. Michael Anthony, Alaska Science Center, National Biological Survey, 1011 East Tudor Road, Anchorage, AK 99503; telephone 907-786-3508; FAX 907-786-3636; E-mail mike_anthony.nbs.gov. Literature cited BAJZAK, D., AND J. F. PIATT. 1990. Computer-aided procedure for counting waterfowl on aerial photographs. Wildl. Soc. Bull. 18: 125-129. GILMER, D. S., J. A. BRASS, L. L STRONG, AND D. H. CARD. 1988. Goose counts from aerial photographs using an optical digitizer. Wildl. Soc. Bull. 16:204-206. HARRIS, M. P., AND C. S. LLOYD. 1977. Variations in counts of seabirds from photographs. British Birds 70:200-205. HEYLAND, J. D. 1972. Vertical aerial photography as an aid in wildlife population studies. Pages 121 - 136 in D. White, editor. Proc. of the first Canadian symposium on remote sensing. Dep. Energy, Mines, and Resour., Ottawa, Ont. KADLEC, J. A., AND W. H. Drury 1968. Aerial estimation of the size of gull breeding colonies. J. Wildl. Manage. 32:287-293. KERBES, R. H. 1983. Lesser snow goose colonies in the western Canadian arctic. J. Wildl. Manage. 47:523-526. LEONARD, R. M., AND E. FISH. 1974. An aerial photographic technique for censusing lesser sandhill cranes. Wildl. Soc. Bull. 2:191-195. Software Editor: Smith