Feature hashing

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

Code Consistent Hashing Based on Information-Theoretic Criterion

Abstract: Learning based hashing techniques have attracted broad research interests in the Big Media research area. They aim to learn compact binary codes which can preserve semantic similarity in the Hamming embedding. However, the discrete constraints imposed on binary codes typically make hashing optimizations very challenging. In this paper, we present a code consistent hashing ( CCH ) algorithm to learn discrete binary hash codes. To form a simple yet efficient hashing objective function, we introduce a new code consistency constraint to leverage discriminative information and propose to utilize the Hadamard code which favors an information-theoretic criterion as the class prototype. By keeping the discrete constraint and introducing an orthogonal constraint, our objective function can be minimized efficiently. Experimental results on three benchmark datasets demonstrate that the proposed CCH outperforms state-of-the-art hashing methods in both image retrieval and classification tasks, especially with short binary codes.

A deep neural network based hashing for efficient image retrieval

Abstract: Learning valid similarities is a vital problem in hashing methods, especially in large-scale image search. Similarity pertained hashing method is widely used in image retrieval for its high quality compact binary code mapping. The hashing scheme of most existing hashing methods is that the input data is encoded as a vector of visual features and hashed into binary hash codes via projection functions or quantization methods afterward. However, this separated pipeline may prone to lose accurate similarities of images, since the limited compatible domain between visual feature vectors generation and binary codes mapping process. Encouraged by the extraordinary image representation learning ability of deep neural networks in classification, we propose a structure that merges binary code generation process within deep neural networks for efficient image retrieval. The proposed architecture contains two fundamental blocks. The stacked convolution layers of Network In Network with global average pooling compute effective image representation and the embedded latent layer with binary activation functions learn binary hash codes simultaneously. Experiments show that the proposed method gains improvement over several state-of-the-art hashing methods.

Partial match retrieval using recursive linear hashing

Abstract: Recursive linear hashing is a hashing technique proposed for files which can grow and shrink dynamically. The scheme is an extension of linear hashing, a method originally proposed by Litwin, but unlike Litwin's scheme, it does not require conventional overflow pages. In this paper, we investigate the application of recursive linear hashing to partial match retrieval problems. Consistent with the results for primary key retrieval, recursive linear hashing performs better than the conventional scheme on these problems, especially at high load factors.

Robust Image Hash Function Based on Polar Harmonic Transforms and Feature Selection

Abstract: Robust hashing aims at representing the perceptual essence of media data in a compact manner, and it has been widely applied in content identification, copyright protection, content authentication, etc. In this paper, a robust hashing algorithm is proposed by incorporating the polar harmonic transforms and feature selection. The proposed hashing algorithm starts by preprocessing, where morphological operations are employed to disclose the principle structures of the input image. The polar harmonic transforms, which have shown promising results in pattern classification, are then exploited to produce candidate features for hash computation. In order to select the most robust and discriminative features, feature selections are applied on the candidate feature set via boosting algorithm. The hash string is finally generated by randomly permuting the quantization indexes of selected features. Experimental results reveal that the proposed work is both distortion-resistent and discriminative, and it can achieve higher content identification accuracy than the comparative algorithm.

A memorability based method for video hashing

Abstract: In this paper, a memorability feature based-video hashing is proposed as an alternative to appearance feature and visual attention based-algorithms. Inspired by our previous study which shows spatial histograms based on visual attention have positive influence on predicting image memorability, we propose to define the memory feature (MF) and use it to characterize the video content. The saliency map (SM) is constructed by the visual saliency detection in video segments and three local visual attention regions are detected from SM. The spatial histograms feature of the visual attention regions are obtained, which are defined as memory feature (MF) and the supervised hashing with kernels (KSH) is adopted to map the MF to hash. Experiments on different kinds of videos demonstrate that the proposed MF-Hashing algorithm is promising in video hashing.