3D invariants with high robustness to local deformations for automated pollen recognition
read more
Citations
Fast and robust segmentation of spherical particles in volumetric data sets from brightfield microscopy
Fast computation of 3D spherical Fourier harmonic descriptors - a complete orthonormal basis for a rotational invariant representation of three-dimensional objects
3D invariants for automated pollen recognition
A Local Feature Descriptor Based on SIFT for 3D Pollen Image Recognition
Analysis of Relevant Features for Pollen Classification
References
An experimental comparison of min-cut/max- flow algorithms for energy minimization in vision
Snakes, shapes, and gradient vector flow
An Experimental Comparison of Min-cut/Max-flow Algorithms for Energy Minimization in Vision
Automatic recognition of biological particles in microscopic images
Invariant Features for Gray Scale Images
Related Papers (5)
Frequently Asked Questions (14)
Q2. Why is the HI framework used in this paper?
Due to the sparsity of non-zero-values in the synthetic channels the resulting integral features are highly localized in the real space, while the framework automatically guarantees the desired invariance properties.
Q3. What is the important requirement for a fully automated system?
A few such pollen grains per m3 of air can already cause allergic reactions) the avoidance of false positives is one of the most important requirements for a fully automated system.
Q4. How can the authors achieve robustness to local deformations?
While the authors achieve full invariance to radial deformations by full Haar-integration the authors can only reach robustness to local deformations by partial Haar-integration.
Q5. What is the first step in processing the pollen monitor data set?
The first step in processing the pollen monitor data set is the detection of circular objects with voxel-wise vector based gray-scale invariants, similar to those in [8].
Q6. What was the training set for the SVM?
From the training set only the “clean” (not agglomerated, not contaminated) pollen and the “non-pollen” particles from a few samples were used to train the support vector machine (SVM) using the RBF-kernel (radial basis function) and the one-vs-rest multi-class approach.
Q7. What is the qr to the segmentation border?
For the application on the pollen monitor data set (rotational invariance only around the z-axis), q is split into a radial distance qr to the segmentation border and the z-distance to the central plane qz.
Q8. What is the sampling of the parameter space of the kernel functions?
The best sampling of the parameter space of the kernel functions (corresponding to the inner class deformations of the objects), was found by cross validation on the training data set, resulting in Nqr ×Nqz ×Nc×n = 31×11×16×16 = 87296 “structural” features (using kernel function k1) and 8 “shape” features (usingkernel function k2).
Q9. What is the main purpose of the project?
Within the BMBF-founded project “OMNIBUSS” a first demonstrator of a fully automated online pollen monitor was developed, that integrates the collection, preparation and microscopic analysis of air samples.
Q10. How is the HI framework extended to global and local deformations?
This is achieved by creating synthetic channels containing the segmentation borders and employing special parameterized kernel functions.
Q11. How many airborne particles were recorded in the confocal data set?
The “pollen monitor data set” contains about 180,000 airborne particles including about 22,700 pollen grains from air samples that were collected, preparedand recorded with transmitted light microscopy from the online pollen monitor from March to September 2006 in Freiburg and Zürich (fig. 1c).
Q12. What is the resulting transformation of the radial deformations?
the group of arbitrary deformations GD and the group of rotations GR the final Haar integral becomes:T = ∫GR∫Gγ∫GDf ( gRgγgDS, gRgγgDX ) p(D) dgD dgγ dgR , (8)where p(D) is the probability for the occurrence of the local displacement field D. The transformation of the data set is described by (gX)(x) =: X(x′), wherex′ = Rx︸︷︷︸ rotation + γ(Rx)︸ ︷︷ ︸ global deformation+
Q13. What is the simplest way to compute a spherical-harmonics?
For 3D rotations this framework uses a spherical-harmonics series expansion, and for planar rotations around the z-axis it is simplified to a Fourier series expansion.
Q14. How can a pollen expert be able to detect the difference between the two samples?
As most pollen grains are nearly spherical and the subtle differences are mainly found near the surface, a pollen expert needs the full 3D information (usually by “focussing through” the transparent pollen grain).