This paper proposes a novel approach for geodesic saliency detection, which focuses on background priors rather than object appearance contrast. The method exploits two common background priors: boundary and connectivity priors. The boundary prior suggests that most image boundaries are background, while the connectivity prior indicates that background regions are usually large and homogeneous. These priors are used to define geodesic saliency, which measures the shortest path from an image patch to a virtual background node. This approach is intuitive, easy to interpret, and allows fast implementation. It is complementary to previous methods that rely on contrast priors, as it benefits more from background priors. Evaluation on two databases shows that geodesic saliency achieves superior results, outperforming previous approaches in both accuracy and speed (2 ms per image). The method is particularly effective in highlighting entire objects uniformly and is more robust to background clutter. Two variants of the algorithm are proposed: GS Grid, which is faster and uses rectangular patches, and GS Superpixel, which is more accurate and uses superpixels. The GS Grid algorithm is significantly faster and suitable for real-time applications, while GS Superpixel is more accurate but slower. The method is validated on the MSRA and Berkeley-300 databases, showing that it outperforms previous methods in both databases. The results indicate that appropriate prior exploitation is helpful for the ill-posed saliency detection problem. The method is also effective in complex scenarios, such as images with multiple foreground objects and complex backgrounds. However, it may fail in cases where objects significantly touch the image boundary or when the background is complex. The paper concludes that further research is needed to improve the method by reducing its dependency on background priors and enhancing its performance in challenging scenarios.This paper proposes a novel approach for geodesic saliency detection, which focuses on background priors rather than object appearance contrast. The method exploits two common background priors: boundary and connectivity priors. The boundary prior suggests that most image boundaries are background, while the connectivity prior indicates that background regions are usually large and homogeneous. These priors are used to define geodesic saliency, which measures the shortest path from an image patch to a virtual background node. This approach is intuitive, easy to interpret, and allows fast implementation. It is complementary to previous methods that rely on contrast priors, as it benefits more from background priors. Evaluation on two databases shows that geodesic saliency achieves superior results, outperforming previous approaches in both accuracy and speed (2 ms per image). The method is particularly effective in highlighting entire objects uniformly and is more robust to background clutter. Two variants of the algorithm are proposed: GS Grid, which is faster and uses rectangular patches, and GS Superpixel, which is more accurate and uses superpixels. The GS Grid algorithm is significantly faster and suitable for real-time applications, while GS Superpixel is more accurate but slower. The method is validated on the MSRA and Berkeley-300 databases, showing that it outperforms previous methods in both databases. The results indicate that appropriate prior exploitation is helpful for the ill-posed saliency detection problem. The method is also effective in complex scenarios, such as images with multiple foreground objects and complex backgrounds. However, it may fail in cases where objects significantly touch the image boundary or when the background is complex. The paper concludes that further research is needed to improve the method by reducing its dependency on background priors and enhancing its performance in challenging scenarios.