Precise Image Exploration With Cluster Analysis

Since the rise of digital multimedia in our present age, when looking for an image that closely matches their needs and preferences, the number of images a user must sort through has become more and more unmanageable. Even when searching for a narrow topic, it can be nearly impossible to find an image that meets a specific preference by going through all the possible images. To combat this growing problem, we describe an exploration system built on deep neural networks to empower the users to quickly sort through all the possible images by quickly narrowing down to their preferred images. By design, our exploration system goes around the need to match the user’s query directly to a small group of images to serve users images that would traditionally be too difficult to group together and match to a query. We propose to use deep metric learning and clustering to group the images, which we will see cleverly manages problems that hold back traditional neural networks in this problem—unseen image groups and shifting definitions