To quickly and efficiently analyze a large-scale environment by the camera with limited field-of-view, intelligent systems should sequentially select the optimal field-of-view to observe important and informative parts of area. Especially in the image retrieval tasks, small observations could be sequentially selected to improve the performance of image retrieval with less computational costs than whole observations at once and the enhanced retrieval performance could be used to select the next best-view again in a cyclic process. In this paper, we have investigated the effects of selected image patches, which might be either overlapped with a certain ratio or non-overlapped with previous observations, in this cyclic process. The adaptive patch selection algorithm is also described as follows: (1) A current observation is decided by its own information gain model which is designed by a similarity value between current observed information and training dataset. (2) After then, the system will update the information gain model by discarding the irrelevant training data with the current observation. During this process, we have shown that an informative patch, even though a part of selected patch is already observed at previous steps, can enhance the retrieval accuracy and it has a better performance than an independent observation method. Experimental results also have shown that the model selects the informative patches around the important contents to retrieve the target images such as the sky, building and so on.
Zhihao Shen, Sungmoon Jeong, Hosun Lee, and Nak Young Chong, Informative Sequential Patch Selection for Image Retrieval, Proceedings of the 2017 IEEE International Conference on Information and Automation (ICIA), Macau SAR, China, 18-20 July 2017 (Candidated to Best Paper Award).