Abstract:The most striking success of deep hashing for large-scale image retrieval benefits from its powerful discriminative representation of deep learning and the attractive computational efficiency of compact hash code learning. Most existing deep semantic-preserving hashing regard the available semantic labels as the ground truth for classification or transform them into prevalent pairwise similarities. However, such strategies fail to capture the interactive correlations between the visual semantics embedded in images and the given category-level labels. Moreover, they utilize the fixed piecewise or pairwise semantics as the optimization objectives, which suffers from the limited flexibility on semantic representation and adaptive knowledge communication in hash code learning. In this paper, we propose a novel Deep Collaborative Graph Hashing (DCGH), which collectively considers multi-level semantic embeddings, latent common space construction, and intrinsic structure mining in discriminative hash codes learning, for large-scale image retrieval. To the best of our knowledge, this is the first collaborative graph hashing for image retrieval. Specifically, instead of using the conventional single-flow visual network architecture, we design a dual-stream feature encoding network to jointly explore the multi-level semantic information across visual and semantic features. Moreover, a well-established shared latent space is constructed based on space reconstruction to explore the concurrent information and bridge the semantic gap between visual and semantic space. Furthermore, a graph convolutional network is introduced to preserve the latent structural relations in the optimal pairwise similarity-preserving hash codes. The whole learning framework is optimized in an end-to-end fashion. Extensive experiments on different datasets demonstrate that our DCGH can achieve superb image retrieval performance against state-of-the-art supervised hashing methods. The source codes of the proposed DCGH are available at https://github.com/JalinWang/DCGH .

GKSH: Graph based Image Retrieval using Supervised Kernel Hashing.

Sparse graph based self-supervised hashing for scalable image retrieval

Kernelized Iterative Hashing for Image Retrieval

Graph-based Supervised Discrete Image Hashing

Kernelized sparse hashing for scalable image retrieval

Supervised hashing with kernels

Image Retrieval via Balanced and Maximum Variance Deep Hashing.

Kernelised Supervised Context Hashing

Fast Graph Similarity Search Via Hashing and Its Application on Image Retrieval

Deep Supervised Hashing for Fast Image Retrieval

Online Supervised Sketching Hashing for Large-Scale Image Retrieval

Deep Graph Laplacian Hashing for Image Retrieval

Kernel-Based Supervised Discrete Hashing For Image Retrieval

Large-scale Image Retrieval with Supervised Sparse Hashing

Deep Collaborative Graph Hashing for Discriminative Image Retrieval

Supervised Hashing Using Graph Cuts and Boosted Decision Trees

Deep and Fast: Deep Learning Hashing with Semi-Supervised Graph Construction.

Toward Making Unsupervised Graph Hashing Discriminative.

Semi-Supervised Deep Hashing with a Bipartite Graph.

Discrete Graph Hashing via Affine Transformation

Probabilistic Hypergraph Based Hash Codes for Social Image Search