With the continuous development of mobile devices, the number of images on the Internet increases explosively. Hashing methods solve retrieval problems with large datasets by converting images into binary hash codes. However,the image dataset on the Internet is updating and its data distribution may change as time goes by. In this situation, the retrieval effectiveness of ordinary hashing methods designed for stationary environments will decline. Thus, hashing methods for non-stationary environments are developed to learn from newly arrived data and adapt to new data environments for better retrieval accuracy in non-stationary environments. In this paper, goals of ideal hashing methods for non-stationary environments are proposed. State-of-the-art hashing methods for non-stationary environments are introduced and analyzed for their advantages and disadvantages according to goals. Experiments are presented to show characteristics of these methods. Suggestions for future development of non-stationary hashing are also given at the end of this paper.
|Journal||International Journal of Machine Learning and Cybernetics|
|Early online date||22 Aug 2022|
|Publication status||Published - Dec 2022|
Bibliographical noteThis work was supported in part by the National Natural Science Foundation of China under Grant 61876066 and in part by China Postdoctoral Science Foundation under Grant 2020M672631.
- Non-stationary environments
- Online hashing