A corner-based saliency model
Wirawit Rueopas, Sangsan Leelhapantu, Thanarat H. Chalidabhongse · 2016
Biological-based bottom-up saliency models base their computation on low-level features like intensity, color and orientation which input signals are encoded along the retina, lateral geniculate nucleus (LGN), and primary visual cortex (V1) respectively. Visual attention is then evaluated independently on each feature channel and combined later to achieve a final saliency map. In this work, we propose a model that utilizes medium-level features such as corners, line intersections, and line endings to extract possible locations of figures in natural images. Our result on Toronto dataset [1] indicates that the model is competitive among other state-of-the-art models. We also believe these features are essential for object learning and recognition at later stage of visual processing in human.