Image Component Library (ICL): Utility classes and Functions

TODO!!! The ICLCV package contains functions and classes for region and blob detection and for tracking. However, the most common component is the icl::RegionDetector class, which performs a parameterized connected component analysis on images. Here are some sample application screenshots:

icl-region-inspector GUI

Blobs and Regions

At first, we have to define the terms "Blob" and "Region":

A "Region" is a set of connected pixels. The set of pixels belonging to a single Region has to fulfill a certain criterion of homogeneity (e.g. "all pixels" have exactly the same value"). The term of "connection" must also be defined more precisely in order to avoid misunderstandings. A pixel is connected to all pixels in it's neighborhood, which in turn is given by the 4 pixels (to the left, to the right, above and below) next to the reference pixel. The also very common neighborhood that contains the 8 nearest neighbours is currently not supported. E.g. a connected component analysis yields a list of Regions.

A Blob is a more general local spot in an image. Blobs don't have to be connected completely.

Detection vs. Tracking

We differentiate explicitly between detection and tracking. When regions or blobs are detected in an image, no prior knowledge for the supposed image location is used, i.e. the region/blob is detected in the whole image.
In contrast, blob tracking of blob means, that blobs are tracked in general from one time step to another, i.e the former blob location is used as prior guess for it's location in the current frame (tracking over time).

Different Blob Detection Approaches

The ICLCV package provides some different blob detection and tracking approaches, that are introduced shorty in the following section. For a more detailed insight look at the corresponding class descriptions.

Color Blob Searcher API (template based)
The Color Blob Searcher API is a template based framework which provides a mechanism for a color based blob detection. In this approach each blob is determined by a special color, so blobs with identical colors are mixed together. Main class is the icl::ColorBlobSearcher template.

Region Detection Package
The icl::RegionDetector performs a connected component analysis on images. Regions that can be found must be connected and must show identical gray values (color images can not be processed yet). Commonly the input image of the RegionDetector's detect(...)-method is a kind of feature map that shows only a small number of different gray values (see the class documentation for more detail). The set of detected image regions can be restricted by: (i) a minimal and maximal gray value and (ii) a minimal and maxmial pixel count Note: The algorithm is highly speed-optimized, by using a special kind of self developed memory handling, which avoids memory allocation and deallocation at runtime if possible. Given large images with O(pixel count) regions (e.g. ordinary gray images instead of a feature map) the algorithm may need more physical memory than available. A very common pre-processing function may be ICLFilter/LUT::reduceBits(..). In section Region Detection Example an example is presented.

Position Tracker template class
These approaches above all perform Blob or region detection. The icl::PositionTracker or it's generalized version icl::VectorTracker can be used to tracking the resulting regions or blobs through time.

Utility classes and Functions
In this group some additional support classes and functions are provided

The Curvature Scale Space

The curvature scale space can be used to extract 2D geometry models from regions. Erik: Please add some information here!

icl-corner-detection-css-demo GUI

Region Detection Example

#include <ICLQt/Common.h>
#include <ICLCV/RegionDetector.h>
#include <ICLFilter/ColorDistanceOp.h>

icl::qt::GUI gui;
GenericGrabber grabber;
RegionDetector rd(100,1E9,255,255);
ColorDistanceOp cd(Color(0,120,240),100);

void mouse(const MouseEvent &e){
  if(e.isLeft()){
    cd.setReferenceColor(e.getColor());
  }
}

void init(){
  grabber.init(pa("-i"));
  gui << Draw().handle("image") << Show();
  gui["image"].install(mouse);
}

void run(){
  DrawHandle draw = gui["image"];
  const core::ImgBase *I = grabber.grab();

  draw = I;

  std::vector<ImageRegion> rs = rd.detect(cd.apply(I));
  for(size_t i=0;i<rs.size();++i){
    draw->linestrip(rs[i].getBoundary());
  }
  draw->render();
}
int main(int n,char **v){
  return ICLApp(n,v,"-input|-i(2)",init,run).exec();
}

icl-online-region-detection-demo screenshot