Visual object recognition via hypothesis construction/verification in chromosome analysis

The problem of automatic human chromosome analysis and recognition can be treated as a typical example of the 2D image understanding task. The objects (chromosomes) have to be detached from the background by the segmentation procedures, their descriptions have to be constructed and decision making (assigning particular chromosome to one of the 24 classes) has to be carried out. Here, the sequential multistage decision making process is constructed, based on combination of local and global object descriptions and "from coarse to fine" analysis. Experimental results on human chromosome recognition are presented.<>