An efficient radix trie‐based semantic visual indexing model for large‐scale image retrieval in cloud environment. (14th April 2020)

• Record Type: Journal Article
• Authors: Krishnaraj, N.; Elhoseny, Mohamed; Lydia, E. Laxmi; Shankar, K.; ALDabbas, Omar
• Other Names: Baker Thar guestEditor.; Al‐Jumeily Dhiya guestEditor.; Maamar Zakaria guestEditor.; Tari Zahir guestEditor.
• Is Part Of: Software, practice & experience. Volume 51:Number 3(2021)
• Page Start: 489
• Page End: 502
• Publication Date: 2020-04-14
• Subjects: cloud computing -- indexing -- large‐scale image retrieval -- platform as a service -- radix trie -- semantic data; Computer software -- Periodicals; Computer programming -- Periodicals; Computer programs -- Periodicals; 005.3
• DOI: 10.1002/spe.2834
• Languages: English
• ISSNs: 0038-0644

Abstract

Summary: In recent years, massive growth in the number of images on the web has raised the requirement of developing an effective indexing model to search digital images from a large‐scale database. Though cloud service offers effective indexing of compressed images, it remains a major issue due to the semantic gap between the user query and diverse semantics of large‐scale database. This article presents a radix trie indexing (RTI) model based on semantic visual indexing for retrieving the images from cloud platforms. Initially, an interactive optimization model is applied to identify the joint semantic and visual descriptor space. Next, an RTI model is applied to integrate the semantic visual joint space model for finding an effective solution for searching large‐scale sized dataset. Finally, a Spark distributed model is applied for deploying the online image retrieval service. The performance of the proposed method is validated on two standard dataset, namely, Holidays 1 M and Oxford 5 K in terms of mean average precision (mAP) and processing time under varying dataset sizes. During experimentation, the presented RTI model shows the maximum mAP value of 0.83 under the dataset size of 1000. Similarly, under the sample count of 1000, it is noted that the standalone server requires a maximum of 118 minutes to complete the process, whereas the spark cluster requires a minimum of around only 19 minutes to finish the process. The experimental outcome showed improvement in terms of various measures over the best rivals in the literature.