A smart image retrieval technique has been an increasing demand by the advancements in the field of computer networks and mobile computing. Generally, uploaded images have textual information associated with the owner’s tags. When an input image is given as a search query, its text content is processed to identify the semantics of the image, and the output will be based on the labels stored in it. Traditional hashing techniques are most commonly used to provide high quality search results for labeled images. In this case, hashing codes are processed based on the semantic preserving information obtained from the features of input images using supervised learning methods. But the delivery of high quality pictures without human interaction, using automated annotation among very large scale image databases, is still a continuous research process. This paper focuses on comparing various methods for reducing the semantic gap between low-dimensional and high-dimensional features. It also focuses on content based image retrieval technique (CBIR), with an unsupervised learning method using convolutional Neural Networks (CNN). © 2020, World Academy of Research in Science and Engineering. All rights reserved.