利用情境資訊的影像搜尋

Exploiting Contextual Information for Visual Search

郭盈希(Yin-Hsi Kuo)

指導教授 ： 徐宏民

國立臺灣大學/電機資訊學院/資訊網路與多媒體研究所/博士(2017年)

https://doi.org/10.6342/NTU201702219

摘要

無資料

關鍵字

視覺文字 ； 整合區域特徵 ； 深度特徵 ； 情境資訊 ； 影像搜尋

並列摘要

With the prevalence of capture devices, people are used to share their images and videos on the social media (e.g., Flickr and Facebook). To provide relevant information (e.g., reviews, landmark names, products) for these uploaded media, the need for effective and efficient visual search (e.g., image retrieval, mobile visual search, product search) is emerging. It enables plenty of applications such as recommendation, annotation, and advertisement. The state-of-the-art approaches (visual features) usually suffer from low recall rates because small changes in lighting conditions, viewpoints, or occlusions could degrade the performance significantly. We observe that enormous media collections are along with rich contextual cues such as tags, geo-locations, descriptions, and time. Hence, we propose to exploit different contextual information with the state-of-the-art visual features for solving the above challenges, and are able to improve the retrieval accuracy and provide diverse search results.

並列關鍵字

Bag-of-Words (BoW) ； Vector of Locally Aggregated Descriptors (VLAD) ； Deep features ； Contextual information ； Visual search

參考文獻

[2] R. Arandjelović, P. Gronat, A. Torii, T. Pajdla, and J. Sivic. NetVLAD: CNN architecture for weakly supervised place recognition. In CVPR, 2016.

[3] R. Arandjelović and A. Zisserman. Three things everyone should know to improve object retrieval. In IEEE Conference on Computer Vision and Pattern Recognition, 2012.

[4] R. Arandjelović and A. Zisserman. All about VLAD. In IEEE Conference on Computer Vision and Pattern Recognition, 2013.

[5] H. Azizpour, A. Sharif Razavian, J. Sullivan, A. Maki, and S. Carlsson. From generic to specific deep representations for visual recognition. In CVPR Workshops, 2015.

[6] A. Babenko and V. Lempitsky. Aggregating local deep features for image retrieval. In ICCV, 2015.