The retrieval of visual cultural symbols is an important research field of inheriting and carrying forward Chinese traditional culture in digital way. Generally visual cultural symbols are foregrounds of natural images, so using shape features in image retrieval that needs image segmentation in advance has great advantages. At present, image segmentation is mostly interactive, which is quite subjective, and it also takes a tremendous amount of work to process a large number of images. Therefore this paper proposed an intelligent image retrieval that combined the saliency region segmentation algorithm based on K-means clustering with the improved shape context. It showed good retrieval results on both property and efficiency.