NOTE Szedlmayer et al.: Automated enumeration of age-0 Cynosaon regahs scale circuli 



339 



Figure 3 



Light intensities along a transect 

 taken from a single scale by distance 

 in mm. (A) Light intensities of a 

 single transect. (B) Smoothed data 

 using a nine-point moving average. 

 Dashed line represents a threshold 

 light intensity of 90. Lowering the 

 threshold to include peak Y would 

 eliminate peaks under Z. 



20 



40 60 

 Pixels 



80 



100 



Figure 4 



Light intensities from a section of a scale 

 transect enlarged to illustrate pixel spacing. 

 Circuli spacing changed, but each minima (a, 

 b, and c) were only counted once. Inflection 

 point (d) was detected but not counted as a cir- 

 culus, because it was not a local minimum in 

 a 20-pixel search width. Horizontal lines above 

 letters represent search width size = 20 pixels. 



and visual counts, because background intensity levels 

 change as the transect moves across the otolith or scale. 

 Thus, some areas may be counted incorrectly when 

 they fall below the selected threshold level as shown 

 for a transect across a weakfish scale (Fig. 3). The local- 

 minimum method solves this problem, since identifica- 

 tion of increments is only dependent on adjacent pixel- 

 light intensity levels. The local-minimum method also 

 responds to changes in increment spacing: as increment 

 spacing increases (measurements from one typical scale 

 ranged from 10.3 to 22.9m -6 ) the algorithm moves 

 greater distances (more pixels) along the transect, but 

 does not count more increments until another minimum 

 is detected. For example, in Figure 4, increments a, 

 b, and c were each counted as one circulus despite 

 changes in circuli spacing, but inflection point d was 

 not counted because it was not a local minimum within 

 a 20-pixel search width. 



Increments narrower than the selected search width 

 would cause errors (e.g., microincrement spacing width 

 = 10 pixels, but search width = 20 pixels). However, 

 this problem can be corrected in three ways: (1) reduce 

 the search width; (2) increase the magnification of the 

 scale or otolith, e.g., from 125 to 400; or (3) increase 

 the number of pixels between increments, i.e., increase 

 the resolution of your system as discussed below. In 

 addition, the true limit of counting narrow increments 

 is not the algorithm, but the resolution limit of the light 

 microscope. After projection of the image onto the 

 video monitor, one pixel corresponds to an actual dis- 



tance of about 0.2m -6 (with the light microscope at 

 1000 x). This 0.2m~ 6 size is the maximum theoretical 

 resolution of any light microscope (Eastman Kodak Co. 

 1980). In addition, ". . .the functional limits are in- 

 variably higher than those derived theoretically" (Cam- 

 pana et al. 1987), therefore several pixels may be pre- 

 sent even between the smallest detectable increments. 

 Other advantages of the present system, as well as 

 other systems, include elimination of data entry and 

 associated transcriptional errors, and establishment of 

 repeatable criteria for ageing of fishes. 



A disadvantage of the present system and other 

 similar systems is that clearly defined increments are 

 needed. This was not a problem with weakfish scales 

 because circuli are distinct (Fig. 1), but application to 

 otolith increments may need further refinement. 

 Another disadvantage is that most systems need multi- 

 ple images to complete a transect reading of a single 

 scale or otolith, which may increase processing time 

 and errors. The multiple-image-per-transect problem 

 results from a limiting number of pixels (512 x 512 

 pixels) in our digitizer, such that lower magnifications 

 (25 or 40) that would encompass the entire transect do 

 not contain enough pixels for accurate identification 

 of increments. Systems with greater pixel resolution 

 (e.g., 1024 x 1024) would alleviate this problem, and are 

 now commercially available. 



