Deep Learning based Enhanced Triplet Nework Model for Landmark Classification in Image Retrieval
: Deep learning exist as a successful models for learning, and representing data through semantic method. These are learned as a Section of a classification task. This research work proposes the model of the triplet network, focusing on learning helpful depictions through distance comparisons. Landmark retrieval is a process of restoring a collection of images along its landmarks which is parallel to required images. In the case office reality, studies regarding landmark retrieval concentrates on utilizing the landmarks geometries regards its similarities especially the visual matches. The visual content of social images is of huge diversity in various landmarks, and a few images have similar patterns among various landmarks. At the same time, we noticed that multimodal contents will be there in social images, i.e., visual content and text tags, and landmark will be with its individual characteristic of both of visual as well as the text content. So that the approach on the basis of matching the similar images, may not be highly appealing in this environment. This research work focuses on, whether the visual and text content could be utilized over geographical correlation for landmark retrieval. Particularly, The work for landmark retrieval form an enhanced Triplet Network modal for landmark classification in order to project social image through multimodal content , by a joint model it integrates both landmark classifier and retirement through multimodal contents. The images with geo-tag will normally noted for classifier learning. On the basis of low rank matrix recovery Visual features get refined, and through automatically labeled images it can learn multimodal classification and group sparse. Finally, candidate images are ranked with the consequence of classification as well as semantic consistency between the visual and text content of the combination office. Research on real data visualize the superiority of this approach like comparing with the existing methods.
Paper
Full text
Deep Learning based Enhanced Triplet Nework Model for Landmark Classification in Image Retrieval
Semantic Scholar · Computer Science · 2019
Abstract
: Deep learning exist as a successful models for learning, and representing data through semantic method. These are learned as a Section of a classification task. This research work proposes the model of the triplet network, focusing on learning helpful depictions through distance comparisons. Landmark retrieval is a process of restoring a collection of images along its landmarks which is parallel to required images. In the case office reality, studies regarding landmark retrieval concentrates on utilizing the landmarks geometries regards its similarities especially the visual matches. The visual content of social images is of huge diversity in various landmarks, and a few images have similar patterns among various landmarks. At the same time, we noticed that multimodal contents will be there in social images, i.e., visual content and text tags, and landmark will be with its individual characteristic of both of visual as well as the text content. So that the approach on the basis of matching the similar images, may not be highly appealing in this environment. This research work focuses on, whether the visual and text content could be utilized over geographical correlation for landmark retrieval. Particularly, The work for landmark retrieval form an enhanced Triplet Network modal for landmark classification in order to project social image through multimodal content , by a joint model it integrates both landmark classifier and retirement through multimodal contents. The images with geo-tag will normally noted for classifier learning. On the basis of low rank matrix recovery Visual features get refined, and through automatically labeled images it can learn multimodal classification and group sparse. Finally, candidate images are ranked with the consequence of classification as well as semantic consistency between the visual and text content of the combination office. Research on real data visualize the superiority of this approach like comparing with the existing methods.