Recently, image representations derived by convolutional neural networks(CNN) have achieved promising performance for instance retrieval, and they outperformthe traditional hand-crafted image features. However, most o...Recently, image representations derived by convolutional neural networks(CNN) have achieved promising performance for instance retrieval, and they outperformthe traditional hand-crafted image features. However, most of existing CNN-based featuresare proposed to describe the entire images, and thus they are less robust to backgroundclutter. This paper proposes a region of interest (RoI)-based deep convolutionalrepresentation for instance retrieval. It first detects the region of interests (RoIs) from animage, and then extracts a set of RoI-based CNN features from the fully-connected layerof CNN. The proposed RoI-based CNN feature describes the patterns of the detected RoIs,so that the visual matching can be implemented at image region-level to effectively identifytarget objects from cluttered backgrounds. Moreover, we test the performance of theproposed RoI-based CNN feature, when it is extracted from different convolutional layersor fully-connected layers. Also, we compare the performance of RoI-based CNN featurewith those of the state-of-the-art CNN features on two instance retrieval benchmarks.Experimental results show that the proposed RoI-based CNN feature provides superiorperformance than the state-of-the-art CNN features for in-stance retrieval.展开更多
As the web grows, the massive increase in information is placing severe burdens on information retrieval and sharing. Automated search engines and directories with small editorial staff are unable to keep up with the ...As the web grows, the massive increase in information is placing severe burdens on information retrieval and sharing. Automated search engines and directories with small editorial staff are unable to keep up with the increasing submission of web sites. To address the problem, this paper presents Infomarker an Internet information service system based on open Directory and Zero-Keywond Inquiry. The oPen Directory sets up a net-community in which the increasing netcitizens can each organize a small portion of the web and present it to the others. By means of Zero-Keywond Inquiry, user can get the information he is interested in without inputting any keyword that is often required by search engines. In Infomarker,user can record the web address he likes and can put forward an information request based on his web records. The information matching engine checks the information in the open Directory to find what fits user's needs and adds it to user's web address records. The key to the matching process is layered keywoof mapping. Infomarker provides people with a whole new approach to getting information and shows a wide prospect.展开更多
基金supported by the National Natural Science Foundation ofChina under Grant 61602253, U1836208, U1536206, U1836110, 61672294, in part by theNational Key R&D Program of China under Grant 2018YFB1003205, in part by the PriorityAcademic Program Development of Jiangsu Higher Education Institutions (PAPD) fund, inpart by the Collaborative Innovation Center of Atmospheric Environment and EquipmentTechnology (CICAEET) fund, China, and in part by MOST under contracts 108-2634-F-259-001- through Pervasive Artificial Intelligence Research (PAIR) Labs, Taiwan.
文摘Recently, image representations derived by convolutional neural networks(CNN) have achieved promising performance for instance retrieval, and they outperformthe traditional hand-crafted image features. However, most of existing CNN-based featuresare proposed to describe the entire images, and thus they are less robust to backgroundclutter. This paper proposes a region of interest (RoI)-based deep convolutionalrepresentation for instance retrieval. It first detects the region of interests (RoIs) from animage, and then extracts a set of RoI-based CNN features from the fully-connected layerof CNN. The proposed RoI-based CNN feature describes the patterns of the detected RoIs,so that the visual matching can be implemented at image region-level to effectively identifytarget objects from cluttered backgrounds. Moreover, we test the performance of theproposed RoI-based CNN feature, when it is extracted from different convolutional layersor fully-connected layers. Also, we compare the performance of RoI-based CNN featurewith those of the state-of-the-art CNN features on two instance retrieval benchmarks.Experimental results show that the proposed RoI-based CNN feature provides superiorperformance than the state-of-the-art CNN features for in-stance retrieval.
文摘As the web grows, the massive increase in information is placing severe burdens on information retrieval and sharing. Automated search engines and directories with small editorial staff are unable to keep up with the increasing submission of web sites. To address the problem, this paper presents Infomarker an Internet information service system based on open Directory and Zero-Keywond Inquiry. The oPen Directory sets up a net-community in which the increasing netcitizens can each organize a small portion of the web and present it to the others. By means of Zero-Keywond Inquiry, user can get the information he is interested in without inputting any keyword that is often required by search engines. In Infomarker,user can record the web address he likes and can put forward an information request based on his web records. The information matching engine checks the information in the open Directory to find what fits user's needs and adds it to user's web address records. The key to the matching process is layered keywoof mapping. Infomarker provides people with a whole new approach to getting information and shows a wide prospect.
