Robust Complex Salient Regions

Abstract. The challenge of interest point detectors is to find, in an unsupervised way, keypoints easy to extract and at the same time robust to image transformations. In this paper, we present a novel set of saliency features that takes into account the region inhomogeneity in terms of intensity and shape. The region complexity is estimated at real-time by means of the entropy of the grey-level information. On the other hand, shape information is obtained by measuring the entropy of normalized orientations. The normalization step is a key point in this process. We compare the novel complex salient regions with the state-of-the-art keypoint detectors. The new set of interest points shows robustness to a wide set of transformations and high repeatability. Besides, we show the temporal robustness of the novel salient regions in two real video sequences.

• Type: Conference
• Year: 2007
• Where: IBPRIA
• Authors: Sergio Escalera, Oriol Pujol, Petia Radeva
• Keywords: Complex Salient Regions, IBPRIA 2007, Image, Image Analysis, Salient Regions, State-of-the-art Keypoint Detectors