Pollen Spotting Results.

Column (A) shows two randomly selected training images with pollen present, taken from the original scanned data set. Each image is 1000 x 1000 pixels, 0.23 μm/pixel. The top image (A1) is out of focus and the bottom image (A2) is in focus. Columns (B) and (C) show the result of expert editing on the windows shown in (A). Two different experts marked the pollen pixels by erasing them from the original image. Column (D) shows pollen pixel predictions (in red) made by the pixel classifier after it had been trained on the 637 training examples provided by each of the 11 experts.