This paper proposes a new method for visual segmentation based on fixation points. The method segments the region of interest in two steps: (1) generating a probabilistic boundary edge map combining multiple visual cues, and (2) finding the optimal closed contour around the fixation point in the transformed polar edge map. The paper shows this fixation-based segmentation approach improves accuracy over previous methods, especially when incorporating motion and stereo cues. It also introduces a region merging algorithm to further refine segmentation results. Evaluation on video and stereo image datasets demonstrates mean F-measures of 0.95 and 0.96 respectively when combining cues, compared to 0.62 and 0.65 without.