Automated Enumeration by Computer 

 Digitization of Age-0 Weakfish 

 Cynoscion regalis Scaie Circuii* 



Stephen T. Szedlmayer 



Margaret M. Szedlmayer 



Auburn University. Marine Extension and Research Center 

 4170 Commanders Drive, Mobile, Alabama 36615 



Michael E. Sieracki 



Virginia Institute of Marine Science, 



College of William and Mary, Gloucester Point, Virginia 23062 



There has been extensive use of 

 daily otolith growth increments in 

 age and growth studies of age-0 

 fishes (Campana and Neilson 1985). 

 Recently, the daily ageing method 

 has been extended to scales (Szedl- 

 mayer et al. In press). However, 

 visually counting increments is tedi- 

 ous, time consuming, and subject to 

 human error (Rice 1987). In an ef- 

 fort to automate the counting of in- 

 crements or daily circuii in scales of 

 age-0 juvenile weakfish Cynoscion 

 regalis, a microcomputer-based sys- 

 tem was used to digitize the video 

 image of a scale, store the light 

 intensities from a radial transect, 

 and count circuii. Circuii were also 

 counted visually to verify the accu- 

 racy of the software. Others have 

 used microcomputer-based systems 

 to aid in increment counting (Tzeng 

 and Yu 1988, Thorrold and Williams 

 1989, Karakiri et al. 1989), but to 

 our knowledge the present algo- 

 rithm is the first method that com- 

 pletely automates increment coun- 

 ting with a high degree of accuracy. 



Materials and methods 



Age-0 juvenile weakfish were col- 

 lected from the York River, Virgi- 

 nia (for further collection methods, 



* Contribution no. 1632 of the Virginia Insti- 

 tute of Marine Science, College of William 

 and Mary. 



see Szedlmayer et al. 1990). Fish, 

 50-140mm standard length (n = 

 45), were anesthetized with tricane 

 methanol sulfate (50 mg MS-222/L 

 seawater), and scales removed from 

 just below the midbody lateral line 

 curve. The scales were placed on a 

 glass slide in water and cleaned 

 with a paint brush, then permanent- 

 ly mounted with a methacrylate 

 copolymer, and covered with a glass 

 cover slip (Flo-Texx liquid cover 

 slip, Lerner Lab.). 



For the visual method, scale cir- 

 cuii were counted twice by the same 

 reader, along a central radius from 

 the focus to the edge, on the anter- 

 ior side, at 125 x magnification on 

 an Olympus BH-2 microscope. If 

 counts were not the same, they 

 were counted a third time. Only 1 

 out of 45 required a third count, and 

 for that scale the counts that were 

 the same were used for comparison 

 with automated computer counts. 



For automated counting, scales 

 were digitized using the same mag- 

 nification and radius as visual counts. 

 Scale images were detected by a 

 Ikigami ITC-510 video camera (625- 

 line resolution) mounted on the 

 microscope, digitized by a Matrox 

 PIP-512B image analyzer, and stored 

 in computer memory in a matrix of 

 512 x 512 picture elements (pixels) 

 with 256 gray levels for each pixel. 

 A PC-AT 286 computer with a 10- 

 meg Hz processor, 1-meg RAM, and 



math coprocessor controlled the 

 image analyzer. A 40-megabyte hard 

 disk and a floppy disk drive were 

 used for image and data storage. 

 Once the image was digitized, it 

 was displayed on a Panasonic PM 

 205A video monitor (1000-line reso- 

 lution) and a mouse was used to con- 

 trol the movement of a cursor mark 

 projected on the image. The cursor 

 was then positioned to select two 

 points defining a transect from the 

 focus to the edge of the scale per- 

 pendicular to the circuii (Fig. 1). 

 Light intensities (gray levels) of 

 three transects, each one pixel 

 apart and one pixel wide, were 

 simultaneously stored to the hard 

 disk. Each transect was then ana- 

 lyzed using a Fortran program to 

 identify and count scale circuii. The 

 algorithm applied a moving average 

 (7, 8, 9, and 10 pixel averages were 

 tried) to smooth each transect and 

 then searched for local minima (e.g., 

 10 pixels on either side of the inflec- 

 tion point corresponds to a local 

 minimum within a search width of 

 20 pixels; search widths of 18, 20, 

 22, and 24 pixels were tried). The 

 Fortran counting algorithm com- 

 pared adjacent pixels and deter- 

 mined if light intensity increased or 

 decreased. Subsequently, an incre- 

 ment was counted only after the 

 following two criteria were satis- 

 fied: (1) an inflection point was 

 detected, i.e., a change in light in- 

 tensity from decreasing to increas- 

 ing, and (2) the inflection point was 

 the minimum light intensity within 

 the specified search width. Depend- 

 ing on the scale size, one to five im- 

 ages were needed to complete a 

 scale count (i.e., with smaller scales 

 the complete scale was included in 

 the digitized image, but with larger 

 scales several images were needed 

 at the same magnification to include 

 all circuii from the focus to the 

 edge). The computer counts (aver- 



Manuscript accepted 14 November 1990. 

 Fishery Bulletin, U.S. 89:337-340 (1991). 



337 



