Pre-attentive visual segmentation algorithm for cognitive robots

Pre-attentive segmentation is one of important process for object-based visual perception of cognitive robots. This paper proposes a Bhattacharyya distance based irregular pyramid method for pre-attentive segmentation. This algorithm hierarchically builds each level of the irregular pyramid by accumulating similar and spatially close nodes at the level below, with the result that the final segments emerge in this process as they are represented by single nodes at certain levels. The Bhattacharyya distance is used to estimate the intralevel similarity so as to improve the segmentation accuracy and robustness to noise. The ability to self-determine the number of segments is also developed for the pyramidal accumulation process by proposing an adaptive neighbor search method. Experimental results have shown that the proposed algorithm outperforms existing image segmentation algorithms in terms of segmentation accuracy and robustness to noise.