Short Description
Icy Feature Detector plug-in implements a series of optimized contour and ridge detectors. The filters are steerable and are based on the optimization of a Canny-like criterion. They have a better orientation selectivity than the classical gradient or Hessian-based detectors.Documentation
In the following we describe the approach of the plug-in for the design of 2D feature detectors. These detectors are a class of steerable functions based on the optimization of a Canny-like criterion. In contrast with previous computational designs, this approach is truly 2D and provides filters that have closed-form expressions. It also yields operators that have a better orientation selectivity than the classical gradient or Hessian-based detectors. We illustrate the method with the design of operators for edge and ridge detection. We present some experimental results that demonstrate the performance improvement of these new feature detectors. We propose computationally efficient local optimization algorithms for the estimation of feature orientation. We also introduce the notion of shape-adaptable feature detection and use it for the detection of image corners.
The plug-in was originally implemented François Aguet for ImageJ, and was ported to Icy by Ricard Delgado-Gonzalo and Zsuzsanna Püspöki.
Steerable Edge Detector
The steerable edge detectors are based on gaussian derivates. The complexity of the filters depends on the highest order derivate used, which for the edge detectors are the 5th order derivates of the gaussian. For 5th order steerable edge detectors rotations are computed using 12 pre-calculated base templates. The base templates are convolutions of the original image with 1st, 3rd, and 5th derivates of the gaussian.
Example
1st, 3rd, and 5th order edge detectors. Higher order detectors result in fewer false detections as well as more precise detection of the feature. This is due to the higher signal-to-noise ratio and orientation selectivity of the higher order detectors. The images below show typical templates for the different orders.
|
The following images show the results of filtering a noisy test image with different orders of steerable edge detectors. The higher the order the detector is, the better the result.
 |
 |
|
|
Original image
|
1st order
|
|
 |
 |
|
|
3rd order
|
5th order
|
Steerable Ridge Detector
The steerable ridge detectors are based on gaussian derivates. The complexity of the filters depends on the highest order derivate used, which for the ridge detectors are the 4th order derivates of the gaussian. For 4th order steerable ridge detectors rotations are computed using 8 pre-calculated base templates. The base templates are convolutions of the original image with 2nd and 4th derivates of the gaussian.
Example
2nd and 4th order ridge detectors. Higher order detectors result in fewer false detections as well as more precise detection of the feature. This is due to the higher signal-to-noise atio and orientation selectivity of the higher order detectors. The two images below show typical templates for the two orders.
|
The following images show the results of filtering a dna cryo electron micrograph with a conventional ridge detector as well as 2nd and 4th order steerable ridge detectors. The 4th order steerable detector yields the best result.
 |
 |
|
|
Original image
|
2nd order classic
|
|
|
 |
|
|
2nd order steerable
|
4th order steerable
|
References
M. Jacob, M. Unser, "Design of Steerable Filters for Feature Detection Using Canny-Like Criteria," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 26, no. 8, pp. 1007-1019.
2 reviews on “Feature Detector”