Deep Self-Taught Hashing for Image Retrieval

Abstract

Hashing algorithm has been widely used to speed up image retrieval due to its compact binary code and fast distance calculation. The combination with deep learning boosts the performance of hashing by learning accurate representations and complicated hashing functions. So far, the most striking success in deep hashing have mostly involved discriminative models, which require labels. To apply deep hashing on datasets without labels, we propose a deep self-taught hashing algorithm (DSTH), which generates a set of pseudo labels by analyzing the data itself, and then learns the hash functions for novel data using discriminative deep models. Furthermore, we generalize DSTH to support both supervised and unsupervised cases by adaptively incorporating label information. We use two different deep learning framework to train the hash functions to deal with out-of-sample problem and reduce the time complexity without loss of accuracy. We have conducted extensive experiments to investigate different settings of DSTH, and compared it with state-of-the-art counterparts in six publicly available datasets. The experimental results show that DSTH outperforms the others in all datasets.

A Survey on Image Retrieval Techniques

“…As per the survey made on various Efficient Manifold Ranking (EMR) [1] which reduced the computational time when compared with LSH and SH, ODBTC [2] indexing produced the better average precision rate of 0.779 when compared with other existing methods. DSTH [3] produced better results for unlabeled data when compared with performance using classification Label. The Experimental result of GDBM [4] on 50,000 datasets provided 25% improvement in accuracy which made it to be applicable for medical field.…”

“…Bit Pattern Feature and Color Co-Occurrence Feature are extracted from the bitmap images andcolorquantizers which are decomposed from image during the encoding process of ODBTC. c) Yu Liu et al [3] suggested a Deep Self-Taught hashing (DSTH) method in order to overcome some of the problems faced by existing image retrieval methods. DSTH is proposed to apply deep hashing on datasets without label.…”

“…Hashing Performance is boosted by combining with the deep learning methods. DSTH [3] method supports both supervised and unsupervised method to enhance the accuracy of hashing. Importance of CBIR is increasing in the field of education, classification and retrieval of bioinformatics and also in medical field.…”

An Encrypted Speech Retrieval Scheme Based on Long Short-Term Memory Neural Network and Deep Hashing

“…Cao et al [12] proposed a novel deep hashing model, Deep Cauchy Hashing (DCH), which can generate compact binary hashing codes using cauchy quantization loss to achieve efficient Hamming spatial retrieval. Liu et al [13] proposed a deep self-learning hash algorithm (DSTH), which can generate a set of pseudo-tags by analyzing the data itself, and then use the discriminant deep model to learn the hash function of the new data. Tang et al [14] proposed a new discriminant deep quantization hash (DDQH) method, which introduces the batch normalization quantization (BNQ) module to improve retrieval accuracy and simultaneously generate more discriminative hash codes.…”

“…Since these perceptual hashing-based encrypted speech retrieval methods utilize the already designed speech features, redesigning the speech feature requires a large number of prior knowledge and experiments, and the retrieval performance of algorithms depend largely on the extracted speech feature. Moreover, deep learning is one of the most important breakthroughs in the artificial intelligence has achieved great success in many fields, such as face retrieval [8,9], cross-modal retrieval [10,11,32], image retrieval [12][13][14][15], speech recognition [16,17,[22][23][24], natural language processing [18,19], audio classification [20,21], emotion recognition [25][26][27][28] and sound detection [29][30][31]. As a machine learning method with multiple hidden layers, deep learning can acquire general abstract features by creating neural network models with multiple hidden layers and using a large number of training data.…”