One of the most challenging issues in Computer Vision field is the detection of salient regions in an image. Psychovisual experiments suggest that, in absence of any external guidance, attention is directed to visually salient locations in the image.
Visual Saliency or Saliency mainly deal with identifying fixation points that a human viewer would focus on at the first glance. Visual saliency usually refers to a property of a “point” in an image (scene), which makes it likely to be fixated.
Most models for visual saliency detection are inspired by human visual system and tend to reproduce the dynamic modifications of cortical connectivity for scene perception. In scientific literature Saliency approaches can be subdivided in three main groups: Bottom-up, Top-down, Hybrid.
Our group developed methods that detect salient regions by analyzing the spatial density of local keyoints of interest, such as SIFT (Scale Invariant Feature Detectors). We also developed methods for image resizing based on saliency information.