A content-guided searching algorithm for balloons

Cohen and Cohen's balloon model improves the original active contour models by providing an ability to search for distant contours. Driven by an artificial inflating force, a balloon's searching ability is rather primitive. In previous papers, we have presented several new models that incorporate different cues in their searching algorithm. In this paper, we present a new method that uses local image contents to guide a balloon's searching process. Sha'ashua and Ullman's saliency map method is used to trace the strongest curve that passes through a particular pixel. This locally-derived information is then used to guide a balloon's searching process. Benefiting from both a global model and locally derived cues, the combined model is faster and more robust than the original balloon model.