CMP-BIA tools

Revision as of 10:49, 19 December 2013 by Borda (talk | contribs)

CMP-BIA tools (ImagejJ/Fiji)
Author Jiří Borovec (web)
Maintainer Jiří Borovec (email)
File CMP_BIA-0.2.jar
Initial release 2013
Latest version December 2th, 2013
Category Segmentation

The CMP-BIA tools is a package for ImageJ/Fiji which will perform image segmentation and registration. The package will contain mainly user-friendly plugins fully compatible with ImageJ 1.x (and later also ImageJ 2.x). All source codes are publicly available as Maven project (see the GitHub repository). The API in this package can be also used for further development of other Java/ImageJ features related to Image Processing.

Note, all included methods are mainly related to medical imaging but it can be also used in the fields.

About us

Center for Machine Perception (CMP) is a university research center performing fundamental and applied research in computer vision, robotics, machine learning, pattern recognition, and mathematics.

Biomedical Imaging Algorithms group (BIA) headed by doc. Jan Kybic develops new algorithms for biomedical image processing. We process images from different modalities, such as magnetic resonance, ultrasound, computed tomography, or microscopy. We work in 2D, 3D and 4D. We know how to preprocess the data, how to register, segment, model, reconstruct and classify them. We use techniques from image processing, numerical mathematics, as well as machine learning.


  • SLIC superpixels - is segmentation method for clustering similar regions - superpixels - in given image which are usually used for other segmentation techniques. The only two parameters are average (initial) size of each superpixel and rigidity parameter in range <0,1>.

Coming soon

  • Initial grid size
  • Simultaneous segmentation and registration

jSLIC - superpixel segmentation

The amount of data in medical imaging to be processed is increasing - images in histology can easily have 50.000x50.000 pixels or even more. The segmentation or registration of these large images is very demanding. The complexity of segmentation and registration can be reduced by using superpixels.

Recently, SLIC (Simple Linear Iterative Clustering) was introduced for general images and presented as a powerful intermediate phase for further image segmentation, classification and registration. SLIC is an adaptation of the k-means algorithm for superpixel generation with two important distinctions: (a) the weighted distance measure combines colour (using the CIELAB colour space, which is widely considered as perceptually uniform for small colour distances) and spatial proximity and (b) the search space is reduced by limiting to a region 2Sx2S, proportional to the superpixel size S. The search space reduction has a great impact on the speed of whole algorithm.

We made a Java-based open source implementation jSLIC - the superpixel clustering with better performance than the original SLIC. Moreover, we proposed a different regularisation parameter, which influences the compactness of resulting superpixels and propose a default value r=0.2. The new post-processing step gives more reliable superpixels shapes, with no need of decreasing superpixel size.

How to use the plugin

jSLIC interface

As you can see the Interface to the plugin contains a parameters for superpixel configuration and also its final visualisation.


For the configuration there are only two parameters to be set:

  • Image segmentation - in general it can be seen as an average superpixels size.
  • Regularisation - influence the compactness of estimated superpixels. The range is from 0 (very elastic superpixels) to 1 (superpixels are nearly squares). Experimentally, we set as optimal value 0.2 for most cases.


To show of handle segmentation results we presented a few approaches:

  • Overlap contours - simply draw the contours on resulting superpixels into the image by chosen colour.
  • Export segments as ROIs - all superpixels are exported as polygons into the ROI Manager.
  • Show final segmentation - add one more stack and fill each superpixel by a random colour.
  • Save segmentation into file - export the superpixel segmentation into a text file as segmentation matrix with labels.

Sample of jSLIC segmented histological tissue Sample of jSLIC segmented histological tissue


comming soon...