With the rapid development of multimedia technology, image management and retrieval has become a research hotspot, current image retrieval techniques become helpless because of ignoring the emotional factors. Based on the segmentation for natural language and application of OCC emotion model, it was realized to use natural language to make emotional semantic retrieval for image database. Using 800 natural scenery images download in Baidu channel to establish image database and test, the experimental results show that the developed system achieves good retrieval effect, which can lay a good foundation for semantic retrieval of more types of images and has some practice value.