Angular Deep Supervised Hashing for Image Retrieval

Chang Zhou,Xuyuan Xu,Peter H W Wong,Lai-Man Po,Kin Wai Lau,Mengyang Liu,Wilson Y F Yuen,Kwok Wai Cheung,Yuzhi Zhao

doi:10.1109/access.2019.2939650

Abstract

Deep learning based image hashing methods learn hash codes by using powerful feature extractors and nonlinear transformations to achieve highly efficient image retrieval. For most end-to-end deep hashing methods, the supervised learning process relies on pair-wise or triplet-wise information to provide an internal relationship of similarity data. However, the use of pair-wise and triplet loss function is limited not only by expensive training costs but also by quantization errors. In this paper, we propose a novel semantic learning based hashing method for image retrieval to optimize the deep features structure in the hash space from a perspective of angular view. Specifically, we proposed an angular hashing loss function that explicitly improve intra-class compactness and inter-class separability between features in hash space. Geometrically, angular hashing loss can be regarded as imposing hash constraints on hypersphere manifold. In order to solve the training problem on the multi-label case, we further designed a dynamic Softmax training strategy that can directly train the network using gradient descent method. Extensive experiments on two well-known datasets of CIFAR-10 and NUS-WIDE demonstrate that the proposed Angular Deep Supervised Hashing (ADSH) method can generate high-quality and compact binary codes, which can achieve state-of-the-art performance as compared with conventional image hashing and deep learning-based hashing methods.

Highlights

With the rapid development of social media and smartphones, huge amount of image data is uploaded to the Internet every minute, such as human face and online products
We studied three semantic learning cases: (1) The network is only supervised by Softmax; (2) The network is only supervised by A-Softmax; (3) Supervise the network through joint supervision of the proposed method (ADSH)
The performance gap between deep features and hash codes verifies that Angular Deep Supervised Hashing (ADSH) method is an effective solution for controlling quantization errors

Summary

Introduction

With the rapid development of social media and smartphones, huge amount of image data is uploaded to the Internet every minute, such as human face and online products. Most recent researches in visual search use content-based image retrieval (CBIR) [1] without relying on label and text information. The image with the smallest distance is considered as the most similar image. These handcrafted features are distinctive with low mismatch probability and good for indexing. The high dimensionality of the feature domain makes searching very challenging, especially for large-scale image databases. To address this problem, many hash-based Approximate Nearest Neighbor (ANN) search methods [4]–[7] have been proposed. Since the hash-based approach [8] can encode an image into a compact binary code with similarity preservation, the burden of computation and memory requirements can be reduced

Objectives

Methods

Results

Conclusion