Most previous works on saliency detection are dedicated to conventional images, however, it is becoming more and more vital to obtain visual attention for panoramic images with the rapid development of VR or AR technology. In this paper, we propose a novel automatic salient object detection framework for panoramic images using region growing and fixation prediction model. First, we employ a spatial density pattern detection method using region growing for the panoramic image to roughly extract the proposal objects. Meanwhile, the eye fixation model is embedded into the framework to predict the visual attention, which simulates the human vision system. Then, the previous saliency information is combined by the maxima normalization to get the coarse saliency map. Finally, a geodesic refinement is utilized to obtain the final saliency map. To fairly evaluate the performance of the proposed framework, we build a new high-quality dataset of panoramic images for the public. Extensive evaluations performed on the new dataset (SalPan) show the superiority of the proposed framework than other methods.
View on arXiv