A framework for image annotation and retrieval from largescale image classes

Abstract

With the swift breakthrough in mobile devices and communication technology, colossal volumes of image data in personal and commercial sites are produced every day and are available to the public. These advancements have propelled a greater interest in digital images and for better ways of archiving, annotating, and accessing the images. Precise image retrieval from large-scale image classes needs effective annotation and indexing of images. Hence, there is an escalating demand for effective and efficient image indexing, annotation and retrieval methods. Currently, there are ample Image annotation and retrieval models available. But a detailed survey reveals that this domain needs much more focus and becomes an active research area. The Image annotation and retrieval model works well when the number of image classes is less, but fails when the image class increases. The facts which substantiate this study are, most of the systems encounter more computations and have limited accuracy. The computation complexity is due to the high-dimensional feature vectors of images belonging to large-scale image classes (Hao et al. 2010; Christos et al. 2010; Zhao et al. 2017). So, there is a need for efficient dimensionality reduction methods. The efficiency of the image retrieval from large-scale image classes largely depends on an effective annotation model. Most of the image annotation models still experience the semantic gap between annotating words and the concept of the image (Wang et al. 2017). The relevancy of the query with retrieved images and time complexity of the retrieval process are influenced by the indexing system (Yan X et al. 2017; Maria T and Anastasios, T 2018). Hence, there is a need for efficient indexing and retrieval methods. For these reasons, Dimensionality reduction methods, Image Annotation, anIndexing and Retrieval techniques need more attention and advanced techniques to handle the issues arising from large-scale image classes newline