Simultaneous feature learning and hash coding with deep neural networks

Abstract

Similarity-preserving hashing is a widely-used method for nearest neighbour search in large-scale image retrieval tasks. For most existing hashing methods, an image is first encoded as a vector of hand-engineering visual features, followed by another separate projection or quantization step that generates binary codes. However, such visual feature vectors may not be optimally compatible with the coding process, thus producing sub-optimal hashing codes. In this paper, we propose a deep architecture for supervised hashing, in which images are mapped into binary codes via carefully designed deep neural networks. The pipeline of the proposed deep architecture consists of three building blocks: 1) a sub-network with a stack of convolution layers to produce the effective intermediate image features; 2) a divide-and-encode module to divide the intermediate image features into multiple branches, each encoded into one hash bit; and 3) a triplet ranking loss designed to characterize that one image is more similar to the second image than to the third one. Extensive evaluations on several benchmark image datasets show that the proposed simultaneous feature learning and hash coding pipeline brings substantial improvements over other state-of-the-art supervised or unsupervised hashing methods. * Corresponding author: Yan Pan, email: panyan5@mail.sysu.edu.cn. posed, e.g., [8,9,4,12,16,27,14,25,3]. The existing learning-based hashing methods can be categorized into unsupervised and supervised methods, based on whether supervised information (e.g., similarities or dissimilarities on data points) is involved. Compact bitwise representations are advantageous for improving the efficiency in both storage and search speed, particularly in big data applications. Compared to unsupervised methods, supervised methods usually embed the input data points into compact hash codes with fewer bits, with the help of supervised information.

Local Deep Hashing Matching of Aerial Images Based on Relative Distance and Absolute Distance Constraints

“…[34] that a convolution layer with a larger convolution kernel replaced by a convolution layer with multiple smaller convolution kernels may not only reduce the number of parameters, but also create more nonlinear mappings and enhance the expression ability of the network. In this paper, the VGG-16 (VGG Network with 16 layers) model is treated as the basic framework, and three same VGG-16 network structures are used to comprise a feature extraction network.…”

“…The other is the "one-stage" network form: Lai et al modified a CNNH (Convolutional Neural Network Hashing) algorithm and the proposed DNNH (Deep Neural Network Hashing) algorithm, which incorporate the process of generating the hash code into the network framework, performed the image feature learning and hash function learning simultaneously in the network, and optimized the overall performance through feedback. Compared to the "two-stage" deep hash algorithms, such a method often yields a better effect [34].…”

Deep Discrete Hashing with Self-supervised Pairwise Labels

“…Following [19], we only use the images associated with the 21 most frequent classes. For these classes, the number of images of each class is at least 5000.…”

Deep Center Supervised Hash for Fast Image Retrieval

“…Due to the advantages of speed and storage, many of the hashing methods [9][10][11][12][13][14][15][16][17][18][19][20][21] have been proposed to raise performance of approximate nearest neighbor search. Data-independent hash methods use unlabeled data for learning a set of hash functions.…”

“…In order to solve this limitation, CNNs-based hash methods [9,10,[18][19][20][21][22][23] is proposed. [9,10,19,22,23] guides the network to learn the approximate binary code output that preserves the semantics similar and dissimilar relation of input image triplets. [18] first decomposes the semantic similarity matrix of the training image data into approximately hash codes, and then uses these approximate hash codes and image labels to train a deep convolution network.…”