Automated multi-day tracking of mice for the analysis of social behavior

where Th is the threshold. The foreground image is composed of all foreground pixels. The threshold is computed with the help of the user, who is prompted to place ellipses on the mice that are visible in 7 randomly selected frames. Using this information, the optimal threshold Th for background subtraction is computed by minimizing a cost function that counts false alarms (the number of pixels outside known mice positions) and misses (number of pixels that do not pass the threshold inside the known mice positions).