A comprehensive study on deep learning approach for CBIR

Abstract

The content analysis of multimedia is used in computer applications. The multimedia data mostly contain digital images. Over the years, the multimedia contents have become more complex. Especially the images, have shown exponential jump. Twitter, Facebook, Instagram and other different archives are flooded with more than millions of images every day. Searching for the right image from the archive is a difficult research task for the computer vision technology. Over the past two decades, there has been an increase in the research area of content-based image retrieval (CBIR), analysis and image classification using deep learning methods. The deep learning methods proved to be an alternative for manual feature engineering where hand-crafted features were created depending on visual contents like shape, color, texture and were adopted in early days. Deep learning is capable of learning the features from the data automatically. Research done in the field of CBIR has shown that, there remained a remarkable gap between the features presented and the visual perception of humans. Researchers focused on reducing this semantic gap to enhance the efficiency of CBIR. This paper shows a decade review of the recent development in CBIR using deep learning methods is done. We will do the performance analysis using the state-of- the-art methods. It will help in future development and research growth in deep learning image retrieval systems.