The invention provides a saliency detection method based on path bottleneck analysis, and belongs to the technical field of image processing. The method comprises four steps of undirected graph expression based on super-pixel segmentation, analysis of the smoothest path, calculation of path bottleneck distance and generation of a saliency map. After the smoothest path between any nodes in an undirected graph and the bottleneck distance on any path are obtained, difference between the super-pixels in an input image can be estimated by utilizing the bottleneck distance on the smoothest path. The semantic relation between image blocks can be estimated, and salient objects with irregular shapes or different scales can also be processed. Compared with the existing saliency detection algorithms, the saliency map generated by the method is more uniform and consistent and higher in accuracy.