Track Objects ( IATrackObjects )

Tracks segments of a segmented image (A) from frame to frame. Additional intensity information in images (B) are used to distinguish segments.
Image A has to be segmented first with the Simple or Watershed Segmentation functions. These segmentation functions segment each frame independently. Tracking will find the instances of the same object in the successive time frames, and mark them as the same segment. The result is a series of modified segmented images where the markers (the values associated with pixels) refer to the same tracked segment throughout the time lapse.
Algorithm: The only available algorithm is Simulated Annealing.
The algorithm randomly changes connections between segments in consecutive frames seeking for the best matching in the vicinity given by the "Maximal link distance". The best match is determined by energy minimization: the squared distance (square of the difference) of: the distance of the center and the size of the segment; and the mean and sum intensities of the associated images (Image Bs) are calculated and normalized by the mean minimal squared distance of these qualities, multiplied by the "Weight factors". The energy of a disconnected segment is 1 for each quality.
To track fusion and fission(division) events - or more often segmentation artifacts - segments can be linked to group of segments. Within "Group size diameter" distance segments can form a group of two if their total area is less then "Maximal group area" and the normalized squared distance of intensities is less than "Group max energy". Grouping is then recursed by "Group recursion" times yielding more complex groups.
Tracking is performed randomly changing connections between segments favoring lower energy states. Initially "thermal noise" is introduced given by the "Starting Temperature" and then it is exponentially decreased by "Temperature divisor" resulting "freezing" of the system in a low energy state.
Segments finally tracked along the optimized links. "Fusion allowed" tracking backwards, thus spatially distant objects that fuse later on will be identified as the same object.
Segments that are represented in less than "Remove segments represented less than" frames are completely removed.
The algorithm is a modification of the tracking algorithm published by Racine et al. Biomedical Imaging: Nano to Macro, 2006. 3rd IEEE International Symposium on Volume , Issue , 6-9 April 2006 Page(s):1020 - 1023.
Hints:
Up to 3 different intensity images can be associated. These can be processed images which can provide texture information e.g. by using high pass filtering, or motion information by using Optical Flow images.
Wider grouping criteria can drastically increase memory requirement and computation time.