Abstract
This paper presents an analysis of using Faiss[8], an open source similarity search library developed at Facebook research, and applying it for the purpose of content based image retrieval(CBIR) using high dimensional feature vectors. Faiss provides novel indexing methods which we have applied for image retrieval applications. Two image descriptors, namely 3-D histograms in HSV color space and convolutional neural network(VGG16) have been used. A third descriptor, which is the concatenation of the two has also been used. This paper shows how the use of Faiss for indexing has significantly reduced the query time with only marginal loss in accuracy. We investigate the use of two different types of image descriptors as mentioned above and how the performance on retrieval differs before and after using Faiss for indexing on some popular datasets.
Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
