Semantic image analysis is an active topic of research in computer vision and pattern recognition. In the last two decades, a large number of works on semantic image analysis have emerged, among which the multi-instance learning (MIL) is one of the most commonly used methods due to its theoretical interest and its applicability to real-world problems. However, compared with various MIL methods and their corresponding applications in the field of semantic image analysis, there is a lack of surveys or review researches about MIL related studies. So the current paper, to begin with, elaborates the basic principles of multi-instance learning, subsequently summarizes it with applications to semantic based- image annotation, image retrieval and image classification as well as several other related applications comprehensively. At length, this paper concludes with a summary of some important conclusions and several potential research directions of MIL in the area of semantic image analysis for the future.