Clustering Based Unsupervised Hashing Framework for Image Retrieval

Abstract

Abstract-This work presents a novel clustering-based unsupervised deep hashing framework for image retrieval that can incrementally train for clustering network and output hash codes while keeping the architecture of the input data distributions, solving the aforementioned problems. The study's deep unsupervised hashing framework retains binary codes' spatial structure without pre-computation of pseudo labels, unlike k-means-based hashing. Clustering-based unsupervised hashing, a deep end-to-end network, may tackle big datasets by optimizing all objective functions concurrently using stochastic gradient descent and mini-batches. Our solution beats state-of-the-art algorithms in extensive CIFAR-10 and NUS-WIDE dataset experiments.