Pollen Detection

Automated Pollen Detection in Multifocal Optical Microscopy Images

Graduation project in collaboration with Dr. Emile A. Hendriks
Information and Communication Theory Group, Technical University Delft.

Sander Landsmeer, Berend Stoel


Figure 2cIn a previous pilot study the feasibility of automated pollen recognition was explored. The overall goal is detecting and recognizing different pollen types, so that a computer aided pollen forecast can be developed. In this project, the aim was to develop the first step in the recognition process; the detection of pollen grain from microscopy images.


  • To develop methods for detecting pollen grains, based on color and shape information, independently of staining variations
  • To combine the information from different focal planes to make the detection method more robust
  • To optimize the method based on a training set, and evaluate the detection in a separate test set


  • Air samples were taken during May and August 2006, resulting in 61 image stacks, divided in a training and test set
  • The detection method comprised of the following steps:
    • Sub-sampling and image registration, to account for small displacements
    • Color similarity transform, to yield a resemblance measure of characteristic color of pollen, stained by Safranin
    • Circle detection, using an adapted Hough transform
    • Clustering candidate pixels from the different focal planes
  • The method was validated using the F2 –measure, which is the weighed harmonic mean of the recall and precision. Additionally the robustness against parameter changes was assessed by the stability of the F-measure at different parameters
  • Parameter settings were optimized using the above figures of merit


  • Recall (% pollen detected correctly) was 86%
  • Precision (positive predicted value, % detected object that were truly pollen) was 61%




Landsmeer SH, Hendriks EA, de Weger LA, Reiber JH, Stoel BC, Detection of pollen grains in multifocal optical microscopy images of air samples. Microsc Res Tech. 2009 Jun;72(6):424-30


Example image of pollen grains
Figure 1.
Example of a microscopic image of an air sample containing pollen grains.

 Figure 2a

 Figure 2b

 Figure 2c

Figure 2 (a) Detection based on circular shape (b). Color similarity of detected objects (c). Combining color and shape properties: circular non-pollen objects are removed


For further information, please contact:
B.C. Stoel, PhD.
Division of Image Processing
Department of Radiology, 1-C2S
Leiden University Medical Center
P.O. Box 9600
2300 RC Leiden
The Netherlands
Tel. +31 (0)71 526 1911
Fax. +31 (0)71 526 6801
e-mail: B.C.Stoel@lumc.nl