Deep Unsupervised Hashing with Latent Semantic Components

Deep unsupervised hashing has been appreciated in the regime of image retrieval. However, most prior arts failed to detect semantic components and their relationships behind images, which makes them lack discriminative power. To make up defect, we propose a novel Semantic Components Hashing (DSCH), involves common sense that an normally contains bunch with homology co-occurrence relationships. Based on this prior, DSCH regards as latent variables under Expectation-Maximization framework designs two-step iterative algorithm objective maximum likelihood training data. Firstly, constructs component structure by uncovering fine-grained semantics images Gaussian Mixture Modal~(GMM), where is represented mixture multiple components, are exploited. Besides, coarse-grained discovered considering between hierarchy organization then constructed. Secondly, close centers at both levels, also share similar each other. Extensive experiments three benchmark datasets demonstrate proposed hierarchical indeed facilitate model achieve superior performance.