An Improved Content Based Image Retrieval System using Unsupervised Deep Neural Network and Locality Sensitive Hashing

Nowadays, many types of digital devices generate data in the form of text, image and video. Hence, it is essential to have a data archival system that maintains and accesses the data. In this context, Content Based Image Retrieval (CBIR) is the problem of searching for images in a large database of images. The CBIR involves feature extraction and identification of similar images. Existing works have utilized supervised deep neural networks to learn the features and used exhaustive searching technique to identify the similar images. However, since real datasets of images are unlabelled and large in size, the usage of supervised deep neural networks and exhaustive searching technique will be a time-consuming process. Thus, this paper proposes an Improved CBIR system that first utilizes an unsupervised deep learning approach to extract the features. Further, it utilizes Locality Sensitive Hashing (LSH) to reduce search space and K Nearest Neighbor (KNN) algorithm on the reduced search space to find the required number of similar images.