I did further processing on a 80x40 sub-image covering the anomalous pixels:
All images are seen at a browser based 8X zoom scale (FF will dither it smoothly, which makes the lines less sharp, but almost as clear).
The information in pixel x/y, RGB & 'gray scale difference' between that pixel and the surrounding pixels was calculated and shown in a sortable table (of 3200 rows).
Each titled section shows an image, as well as the first six rows of the table, indicating both the sort keys used, and the rows selected.
Image of interest - nothing selected.

The 3 darkest pixels seem obvious, without any further prompt..
X Y RGB Diff. Select
0 0 192 20.75 -
1 0 186 21.625 -
2 0 174 25.625 -
3 0 177 23.375 -
4 0 175 24.375 -
5 0 181 24.25 -
6 0 189 19.75 -
Darkest Pixels

But here are the darkest pixels high-lit with green cross-hairs just to check our logic.
X Y RGB ▲ Diff. Select
16 25 0 240.625 TRUE
6 30 0 240.125 TRUE
74 7 6 230.875 TRUE
78 31 112 37.875 -
79 34 112 22.75 -
75 37 115 31.375 -
Next Darkest Pixels

The difference in gray scale between the 'next darkest' pixels and their immediate neighbors is much less obvious. Not nearly as obvious as the darkest 3 pixels.
X Y RGB ▲ Diff. Select
16 25 0 240.625 -
6 30 0 240.125 -
74 7 6 230.875 -
78 31 112 37.875 TRUE
79 34 112 22.75 TRUE
75 37 115 31.375 TRUE
Next Biggest Difference

The next biggest difference in pixels by gray scale is similarly nondescript. The difference drops from over 200 pixels to less than 50.
X Y RGB Diff. ▼ Select
16 25 0 240.625 -
6 30 0 240.125 -
74 7 6 230.875 -
37 18 205 47.875 TRUE
31 20 209 45.375 TRUE
59 13 196 44.875 TRUE
Conclusion
Based on the data it seems those 3 darkest pixels are outliers & anomalous, so I'd say "image glitch" failing a more definitive answer.
Code
See this answer on SO for the Java code used to generate the images. It hot-links to the first, cropped image.