Hashing Method For Image Retrieval Research Articles

Deep Learning-Based Image Retrieval With Unsupervised Double Bit Hashing

Unsupervised image hashing is a widely used technique for large-scale image retrieval. This technique maps an image to a finite length of binary codes without extensive human-annotated data for compact storage and effective semantic retrieval. This study proposes a novel deep unsupervised double-bit hashing method for image retrieval. This approach is based on the double-bit hashing method, which has been shown to better preserve the neighboring structure of binary codes than single-bit hashing. Traditional double-bit hashing methods require the entire dataset to be processed simultaneously to determine optimal thresholding values of binary feature encoding. In contrast, the proposed method trains the hashing layer in a minibatch manner, allowing for adaptive threshold learning through a gradient-based optimization strategy. Additionally, unlike most former methods, which only train the hashing networks on top of fixed pre-trained neural networks backbone. The proposed learning framework trains both hashing and backbone networks alternately asynchronously. This strategy enables the model to maximize the learning capability of the hashing and backbone networks. Furthermore, adopting the lightweight Vision Transformer (ViT) in the proposed method allows the model to capture both local and global relationships between multiple image views exemplar, which lead to better generalization, thus maximizing the retrieval performance of the model. Extensive experiments on CIFAR10, NUW-WIDE, and FLICKR25K datasets validate that the proposed method has superior retrieval quality and computational efficiency than state-of-the-art methods.

Deep debiased contrastive hashing

Hashing has achieved great success in multimedia retrieval due to its high computing efficiency and low storage cost. Recently, contrastive-learning-based hashing methods have achieved decent retrieval performance in label-free scenarios by learning distortion-invariant representations with Siamese networks. Their learning principle, i.e., instance discrimination, maximizes the correlation between self-augmented views and treats all others as negative samples. However, it may learn with false negative samples that are naturally similar, resulting in biased hash learning. To bridge this flaw, we reveal the between-instance similarity of naturally similar samples by exploring the latent structure of the training data. As a result, we propose the Deep Debiased Contrastive Hashing (DDCH) algorithm, using the neighborhood discovery module to explore the intrinsic similarity relationship that can help contrastive hashing reduce false negatives for superior discriminatory ability. Furthermore, we elucidate the rationale for incorporating the module into the contrastive hashing framework and explain our hashing process from an Expectation-Maximization (EM) perspective. Extensive experimental results on three benchmark image datasets demonstrate that DDCH significantly outperforms the state-of-the-art unsupervised hashing methods for image retrieval.

Learning to Hash With Dimension Analysis Based Quantizer for Image Retrieval

The last few years have witnessed the rise of the big data era, in which approximate nearest neighbor search is a fundamental problem in many applications, such as large-scale image retrieval. Recently, many research results demonstrate that hashing can achieve promising performance due to its appealing storage and search efficiency. Since the complex optimization problems for loss functions are difficult to solve, most hashing methods decompose the hash codes learning problem into two steps: projection and quantization. In the quantization step, binary codes are widely used because ranking them by Hamming distance is very efficient. However, the huge information loss produced by the quantization step should be reduced in applications, such as image retrieval where high search accuracy is required. Since many two-step hashing methods produce uneven projected dimensions in the projection step, in this paper, we propose a novel dimension analysis based quantization method (DAQ) on two-step hashing methods for image retrieval. We first perform an importance analysis of the projected dimensions and select a subset of them that are more informative than the others, then we divide the selected projected dimensions into several regions with our quantizer. Every region is quantized with its corresponding codebook. Finally, the similarity between two hash codes is estimated by Manhattan distance between their corresponding codebooks, which is also efficient. We conduct experiments on three public benchmarks containing up to one million descriptors and show that the proposed DAQ method consistently leads to significant accuracy improvements over state-of-the-art quantization methods.

Slice-feature based deep hashing algorithm for remote sensing image retrieval

Remote sensing image retrieval is aimed at obtaining a target image from a typically large number of remote sensing images. Different from natural scene images, remote sensing images usually includes more channels such as near-infrared reflectances (NIR), which will result in higher data dimensionality. However, high dimensionality of image features may lead to information redundancy and high computational costs. Recently, deep hashing methods to reduce image feature dimensionality have been proposed. However, compared with natural images, remote sensing images have smaller interclass distances, which hinder the application of deep hashing methods for image retrieval. To overcome this problem, in this paper, we develop a slice-feature deep hashing (SFDH) method for remote sensing image retrieval. The major contribution of this work is proposing an image correlation reduction strategy by separating the features in the fully connected layers into many slices. The SFDH architecture is based on a pre-trained Inception V3 model; the proposed model not only extracts image features from remote sensing images, but also introduces the slice-feature strategy to improve the fully connected structure of the previously proposed metric learning based deep hashing network (MiLaN). Besides, triplet loss is utilized to further enlarge differences between classes. Our experimental results demonstrate that the proposed slice-feature strategy outperforms state-of-the-art remote sensing image retrieval methods.

Unsupervised semantic deep hashing

In recent years, deep hashing methods have been proved to be effective since it employs convolutional neural network to learn features and hashing codes simultaneously. However, these methods are mostly supervised. In real-world applications, it is a time-consuming and overloaded task for annotating a large number of images. In this paper, we propose a novel unsupervised deep hashing method for large-scale image retrieval. Our method, namely unsupervised semantic deep hashing (USDH), uses semantic information preserved in the CNN feature layer to guide the training of network. We enforce four criteria on hashing codes learning based on VGG-19 model: 1) preserving relevant information of feature space in hashing space; 2) minimizing quantization loss between binary-like codes and hashing codes; 3) improving the usage of each bit in hashing codes by using maximum information entropy, and 4) invariant to image rotation. Extensive experiments on CIFAR-10, NUSWIDE have demonstrated that USDH outperforms several state-of-the-art unsupervised hashing methods for image retrieval. We also conduct experiments on Oxford 17 datasets for fine-grained classification to verify its efficiency for other computer vision tasks.

Generating region proposals for histopathological whole slide image retrieval

Background and objectiveContent-based image retrieval is an effective method for histopathological image analysis. However, given a database of huge whole slide images (WSIs), acquiring appropriate region-of-interests (ROIs) for training is significant and difficult. Moreover, histopathological images can only be annotated by pathologists, resulting in the lack of labeling information. Therefore, it is an important and challenging task to generate ROIs from WSI and retrieve image with few labels. MethodsThis paper presents a novel unsupervised region proposing method for histopathological WSI based on Selective Search. Specifically, the WSI is over-segmented into regions which are hierarchically merged until the WSI becomes a single region. Nucleus-oriented similarity measures for region mergence and Nucleus–Cytoplasm color space for histopathological image are specially defined to generate accurate region proposals. Additionally, we propose a new semi-supervised hashing method for image retrieval. The semantic features of images are extracted with Latent Dirichlet Allocation and transformed into binary hashing codes with Supervised Hashing. ResultsThe methods are tested on a large-scale multi-class database of breast histopathological WSIs. The results demonstrate that for one WSI, our region proposing method can generate 7.3 thousand contoured regions which fit well with 95.8% of the ROIs annotated by pathologists. The proposed hashing method can retrieve a query image among 136 thousand images in 0.29 s and reach precision of 91% with only 10% of images labeled. ConclusionsThe unsupervised region proposing method can generate regions as predictions of lesions in histopathological WSI. The region proposals can also serve as the training samples to train machine-learning models for image retrieval. The proposed hashing method can achieve fast and precise image retrieval with small amount of labels. Furthermore, the proposed methods can be potentially applied in online computer-aided-diagnosis systems.

Deep Quantization Network for Efficient Image Retrieval

Hashing has been widely applied to approximate nearest neighbor search for large-scale multimedia retrieval. Supervised hashing improves the quality of hash coding by exploiting the semantic similarity on data pairs and has received increasing attention recently. For most existing supervised hashing methods for image retrieval, an image is first represented as a vector of hand-crafted or machine-learned features, then quantized by a separate quantization step that generates binary codes. However, suboptimal hash coding may be produced, since the quantization error is not statistically minimized and the feature representation is not optimally compatible with the hash coding. In this paper, we propose a novel Deep Quantization Network (DQN) architecture for supervised hashing, which learns image representation for hash coding and formally control the quantization error. The DQN model constitutes four key components: (1) a sub-network with multiple convolution-pooling layers to capture deep image representations; (2) a fully connected bottleneck layer to generate dimension-reduced representation optimal for hash coding; (3) a pairwise cosine loss layer for similarity-preserving learning; and (4) a product quantization loss for controlling hashing quality and the quantizability of bottleneck representation. Extensive experiments on standard image retrieval datasets show the proposed DQN model yields substantial boosts over latest state-of-the-art hashing methods.

Deep Hashing Network for Efficient Similarity Retrieval

Due to the storage and retrieval efficiency, hashing has been widely deployed to approximate nearest neighbor search for large-scale multimedia retrieval. Supervised hashing, which improves the quality of hash coding by exploiting the semantic similarity on data pairs, has received increasing attention recently. For most existing supervised hashing methods for image retrieval, an image is first represented as a vector of hand-crafted or machine-learned features, followed by another separate quantization step that generates binary codes. However, suboptimal hash coding may be produced, because the quantization error is not statistically minimized and the feature representation is not optimally compatible with the binary coding. In this paper, we propose a novel Deep Hashing Network (DHN) architecture for supervised hashing, in which we jointly learn good image representation tailored to hash coding and formally control the quantization error. The DHN model constitutes four key components: (1) a sub-network with multiple convolution-pooling layers to capture image representations; (2) a fully-connected hashing layer to generate compact binary hash codes; (3) a pairwise cross-entropy loss layer for similarity-preserving learning; and (4) a pairwise quantization loss for controlling hashing quality. Extensive experiments on standard image retrieval datasets show the proposed DHN model yields substantial boosts over latest state-of-the-art hashing methods.

Incremental Hashing for Semantic Image Retrieval in Nonstationary Environments.

A very large volume of images is uploaded to the Internet daily. However, current hashing methods for image retrieval are designed for static databases only. They fail to consider the fact that the distribution of images can change when new images are added to the database over time. The changes in the distribution of images include both discovery of a new class and a distribution of images within a class owing to concept drift. Retraining of hash tables using all images in the database requires a large computation effort. This is also biased to old data owing to the huge volume of old images which leads to a poor retrieval performance over time. In this paper, we propose the incremental hashing (ICH) method to deal with the two aforementioned types of changes in the data distribution. The ICH uses a multihashing to retain knowledge coming from images arriving over time and a weight-based ranking to make the retrieval results adaptive to the new data environment. Experimental results show that the proposed method is effective in dealing with changes in the database.

Supervised Hashing for Image Retrieval via Image Representation Learning

Hashing is a popular approximate nearest neighbor search approach for large-scale image retrieval. Supervised hashing, which incorporates similarity/dissimilarity information on entity pairs to improve the quality of hashing function learning, has recently received increasing attention. However, in the existing supervised hashing methods for images, an input image is usually encoded by a vector of hand-crafted visual features. Such hand-crafted feature vectors do not necessarily preserve the accurate semantic similarities of images pairs, which may often degrade the performance of hashing function learning. In this paper, we propose a supervised hashing method for image retrieval, in which we automatically learn a good image representation tailored to hashing as well as a set of hash functions. The proposed method has two stages. In the first stage, given the pairwise similarity matrix $S$ over training images, we propose a scalable coordinate descent method to decompose $S$ into a product of $HH^T$ where $H$ is a matrix with each of its rows being the approximate hash code associated to a training image. In the second stage, we propose to simultaneously learn a good feature representation for the input images as well as a set of hash functions, via a deep convolutional network tailored to the learned hash codes in $H$ and optionally the discrete class labels of the images. Extensive empirical evaluations on three benchmark datasets with different kinds of images show that the proposed method has superior performance gains over several state-of-the-art supervised and unsupervised hashing methods.