Multi-instance learning(MIL) has been recognized as the fourth machine learning framework, and has been widely
used in the image semantic analysis. Firstly, the concepts such as development history, characteristics and many useful testing
datasets of MIL techniques are reviewed. Then, many popular MIL algorithms are also introduced in detail by using realworld
applications based on image semantic analysis. Meanwhile, based on their machine learning mechanisms, related MIL
algorithms are divided into a variety of categories, which highlights the processes and dominant features of different MIL
algorithms. Finally, the trends and possible outputs for further researches are discussed in details.