EXTRACTING SALIENT CURVES FROM IMAGES - AN ANALYSIS OF THE SALIENCY NETWORK

The Saliency Network proposed by Shashua and Ullman (1988) is a well-k nown approach to the problem of extracting salient curves from images while performing gap completion. This paper analyzes the Saliency Netw ork. The Saliency Network is attractive for several reasons. First, th e network generally prefers long and smooth curves over short or wiggl y ones. While computing saliencies, the network also fills in gaps wit h smooth completions and tolerates noise. Finally, the network is loca lly connected, and its size is proportional to the size of the image. Nevertheless, our analysis reveals certain weaknesses with the method. In particular, we show cases in which the most salient element does n ot lie on the perceptually most salient curve. Furthermore, in some ca ses the saliency measure changes its preferences when curves are scale d uniformly. Also, we show that for certain fragmented curves the meas ure prefers large gaps over a few small gaps of the same total size. I n addition, we analyze the time complexity required by the method. We show that the number of steps required for convergence in serial imple mentations is quadratic in the size of the network, and in parallel im plementations is linear in the size of the network. We discuss problem s due to coarse sampling of the range of possible orientations. Finall y, we consider the possibility of using the Saliency Network for group ing. We show that the Saliency Network recovers the most salient curve efficiently, but it has problems with identifying any salient curve o ther than the most salient one.