There are several distinctive characteristics of these images. In general, the variation is large from experiment to experiment and it is inhomogeneous even within one cell. Due to the potential significance of the results, the results need to be accurate and must also be repeatable. These requirements exclude commonly used techniques for computer vision and image analysis problems. For example, numerous techniques rely on gradient-based methods for optimization and the results heavily depend the initialization and so they can not be repeated in general and thus not applicable.

Here we develop and implement highly automated image analysis methods that are guaranteed to be repeatable. In other words, our solutions utilize only methods that provide global convergence that are repeatable. These methods, however, are more computationally expensive and in certain cases do not admit computationally feasible methods. To overcome these difficulties, we need user interactions. One of the research goals is to minimize user interactions and their effects on the repeatability.

An example of such a system is FISHfinder@FSU.