Feature hashing

About: Feature hashing is a research topic. Over the lifetime, 993 publications have been published within this topic receiving 51462 citations.

Image authentication using global and local features

Abstract: Image authentication is the process of proving image integrity and authenticity. Robust hashing method is developed for detecting image forgery which includes removal, insertion of objects, and abnormal color modification. Global, local and shape features are used in forming the hash. Global features are based on Zernike moments which represent luminance and chrominance characteristics of the image. The local features include position and texture information of salient regions in the image. Shape Feature provides an outstanding description of the geometric structure of shapes. Secret keys are introduced in feature extraction and hash construction. The hash value of a test image is compared with that of a reference image. When the hash distance is greater than or less than a threshold, the received image is judged as a forged image. Collision probability is very low.

Multiple-Feature Kernel Hashing for Mass Detection

Abstract: Medical image retrieval is very important for multimedia applications. Precision of abnormity detection is the key factor of improving retrieval accuracy. This paper proposed a novel detection algorithm based on multiple-feature kernel hashing. To comprehensively describe breast image, different specific features of each suspicious region are extracted, such as CNN deep features, hierarchy weigh Gist features and Histogram of Oriented Gradient (HOG) features. Integrating multiple features and kernel-based supervised hashing algorithm, we achieve an efficient mass detection system for digital mammo grams. Large scale experimental results on Digital Database for Screening Mammography (DDSM) show that the proposed method in this paper can effectively improve the whole detection performance by maintaining the high sensitivity and low false positive rate at the same time.

Random access from compressed datasets with perfect value hashing

Abstract: A representation technique is presented allowing for quick access of individual records from a static compressed dataset. Given a collection of key-record pairs, the representation allows the appropriate short record to be returned for any given key. The approach is a generalization of perfect address hashing. The new approach, called perfect value hashing, uses a carefully chosen pseudo-random number generator to directly produce the correct record for any key in the dataset. This contrasts with address hashing where the random number provides an address which is then used to recover the record from a separate table. Value hashing doesn't have the theoretical limitations of address hashing, and in practice is more space efficient for records of size less than 36 bits. Value hashing has the added benefit (important when the records are encoded for compression) that variable length records can be represented without an increase in the size of the encoded records. This new technique was used to provide random access from a highly compressed spelling dictionary.

A Dynamic Hashing Algorithm Suitable for Embedded System

Abstract: With the increasing of the data numbers, the linear hashing will be a lot of overflow blocks result from Data skew and the index size of extendible hash will surge so as to waste too much memory. This lead to the above two Typical Dynamic hashing algorithm don’t suitable for embedded system that need certain real-time requirements and memory resources are very scarce. To solve this problem, this paper was proposed a dynamic hashing algorithm suitable for embedded system combining with the characteristic of extendible hashing and linear hashing.it is no overflow buckets and the index size is proportional to the adjustment number. DOI: http://dx.doi.org/10.11591/telkomnika.v11i6.2672 Full Text: PDF

IDHashGAN: Deep Hashing With Generative Adversarial Nets for Incomplete Data Retrieval

Abstract: Benefiting from low storage costs and high retrieval efficiency, hash learning has been a widely adopted technology for approximating nearest neighbor in large-scale data retrieval. Deep learning to hash greatly improves image retrieval performance by integrating feature learning and hash coding into an end-to-end framework. However, subject to application scope, most existing deep hashing methods only apply to retrieval of complete data and have undesirable results when retrieving incomplete but valuable data. In this paper we propose IDHashGAN, a novel deep hashing model with generative adversarial networks to retrieve incomplete data, in which feature restoration, feature learning and hash coding are integrated into an unified end-to-end framework. The proposed model consists of four key components: (1) reconstructive and generative loss are used to generate continuous feature of incomplete data in generative network; (2) supervised manifold similarity is proposed to improve retrieval accuracy and obtain good user acceptance; (3) adversarial and classified loss are designed to distinguish authenticity and similarity in discriminative network; and (4) encoding and quantization loss are adopted to preserve similarity and control hash quality. Extensive experiments on benchmark datasets show that IDHashGAN is competitive on complete dataset and yields substantial boosts of 70% on incomplete datasets compared to state-of-the-art hashing methods.