International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 06 Issue: 12 | Dec 2019 www.irjet.net p-ISSN: 2395-0072 © 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 662 “Semantic Retrieval of Trademarks based on Text and Images Conceptual Similarity using Deep Learning” Prof. Pramod Dhamdhere 1 , Ashwini Nilakh 2 , Sushmita Choudhari 3 , Komal Jadhav 4 , Namrata Kate 5 , Ankita Temgire 6 1 Professor, Dept. of information Technology, BSIOTR College, Wagholi, Pune, Maharashtra, India 2,3,4,5,6 UG. Students, Dept of information Technology, BSIOTR College, Wagholi, Pune, Maharashtra, India ---------------------------------------------------------------------***---------------------------------------------------------------------- Abstract - The number of images associated with weakly supervised user-provided tags has increased dramatically in recent years. User-provided tags are inadequate, subjective and noisy. In proposed system, focused on the problem of social image understanding, i.e., tag refinement, tag assignment, and image retrieval. Different from past work, system propose a novel weakly supervised deep matrix factorization algorithm, which uncovers the latent image representations and tag representations embedded in the latent subspace by collaboratively exploring the weakly supervised tagging information, the visual structure, and the semantic structure. The semantic and visual structures are jointly incorporated to learn a semantic subspace without over-fitting the noisy, incomplete, or subjective tags. Besides, to remove the noisy or redundant visual features, a sparse model is imposed on the transformation matrix of the first layer in the deep architecture. Extensive experiments on real world social image databases are conducted on the tasks of image understanding: image tag refinement, assignment, and retrieval. Encouraging results are achieved, which demonstrates the effectiveness of the proposed method. Finally, a unified optimization problem with a well-defined objective function is developed to formulate the proposed problem and solved by a gradient descent procedure with curvilinear search. Extensive experiments on real world social image databases are conducted on the tasks of image understanding: image tag refinement, assignment, and retrieval. Encouraging results are achieved with comparison with the state of-the-art algorithms, which demonstrates the effectiveness of the proposed method. A trademark is a mark that you can use to recognize your business products or services from those of other vendors. It can be represented graphically in the form of any Symbol, logo, words etc. so, they need to be protection. The conceptual similarities among trademarks, which happens when more than two or more trademark similar. Keywords: User Provided Tags, Image Tag Refinement, Image Tag Assignment, Image Tag Retrieval, Social Image Understanding. I) INTRODUCTION In the social media networks human is considered as open and complex framework. The requirements of the user changed likewise because the expectation of one person may subspace by cooperatively investigating the weakly supervised tagging data, semantic structure and visual structure. Recent years have witnessed an increase in the number of community-contributed images associated with rich contextual information such as user-provided tags. These users gave tags can portray the semantic substance of pictures to some degree, which is valuable to numerous tasks, for example, picture tagging (which can be treated as an image to tag search), Content-Based Image Retrieval (CBIR) and Tag-Based Image Retrieval (TBIR). Subsequently, it is vital yet difficult to cooperatively investigate the rich data of network contributed pictures that is regularly normally accessible. By and by, connections are constantly required for numerous tasks, for example, picture tag relationship for cross modular search (i.e., picture tagging and TBIR), picture relationship for CBIR and tag connection for tag extension in true applications, and these connections must be exact. The quantity of pictures related with weakly supervised user-provided tags has expanded significantly as of late. User-provided tags are insufficient, abstract and boisterous. System centre on the issue of social picture understanding, for example tag assignment, image retrieval and tag refinement. System propose a weakly supervised deep matrix factorization algorithm, in which reveals the inactive picture portrayals and tag portrayals inserted in the dormant environment in service is awareness about the circumstance. Then it can be easily adjusted to the dynamic service. II) LITERATURE SURVEY J. Tang et al. [5] present many image processing and pattern recognition problems; visual contents of images are currently de-scribed by high-dimensional features, which are often repetitive and loud. Creators proposed a novel unsupervised component choice plan, to be specific, non- negative phantom investigation with obliged excess, by together utilizing non-negative otherworldly clustering and redundancy analysis. The presented method can directly identify a discriminative subset of the most useful and redundancy-constrained features. Z. Li et al. [6] present performance of TBIR is limited due to incorrect or noisy tag associated with the image uploaded on social websites. To overcome the performance issues some previous image retagging techniques are proposed to fine tune the tag information of social image in transductive learning manner. However, most of the techniques are unable to handle the images which are not part of sampling data. In author proposed an approach of novel factorization called as Projective matrix factorization with unified