Fuzzy Global Segmentation system for video telephony applications

摘要

In this paper a generic, automatic and adaptive video segmentation algorithm is presented. The algorithm is fully automatic, no human interaction is required. It is also generic and adaptive, no a priori knowledge of the scene or camera is assumed. The algorithm works by making the background 'fuzzy' when environmental conditions are unfavorable, then it only transmits the objects of interest, thus respecting constraints such as computational time and ability to produce a segmentation without loss of useful information. Our approach solves the three main issues with recent video segmentations methods, i.e. problems associated with initialization and poor convergence to concave boundaries, the computational cost related to the complexity of the model and the reduction in the number of parameters to be set. The advantage of the proposed approach lies in its simplicity, allowing to define a safety margin around objects of interest by classifying pixels located in the uncertain boundary region, especially when said objects have a cluttered background. Experimental results show that the method is intuitive, easy to control and runs fast.