Abstract:

We proposed a segmentation algorithm for objects that exhibit relatively homogeneous photometric characteristics, embedded in complex background clutter. And we introduced the global minimization of the active contour model into our proposed method so as to get a fast global minimization effect. By our method, a simple binary classifier can be arrived by summing the logprobability of error, then, followed by the statistics of the data of the lookout region. At last, the result of segmentation can be attained by global minimization method. Experimental results show the superior performance of our method in computational complexity and segmentation accuracy compared to that of the popular method of C-V model.

Key words: global minimization, lookout region, missed detections, false alarms, statistics of the data