To Show the Multiple Alignment of the Image search Hash Efficiency

Abstract

Hashing is a popular and efficient method fornearest neighbor search in large-scale data spaces, by embeddinghigh-dimensional feature descriptors into a similarity-preservingHamming space with a low dimension. For most hashing methods,the performance of retrieval heavily depends on the choice ofthe high-dimensional feature descriptor. Furthermore, a singletype of feature cannot be descriptive enough for different imageswhen it is used for hashing. Thus, how to combine multiplerepresentations for learning effective hashing functions is animminent task. In this paper, we present a novel unsupervisedMultiview Alignment Hashing (MAH) approach based on RegularizedKernel Nonnegative Matrix Factorization (RKNMF),which can find a compact representation uncovering the hiddensemantics and simultaneously respecting the joint probabilitydistribution of data. Specifically, we aim to seek a matrixfactorization to effectively fuse the multiple information sourcesmeanwhile discarding the feature redundancy. Since the raisedproblem is regarded as nonconvex and discrete, our objectivefunction is then optimized via an alternate way with relaxationand converges to a locally optimal solution. After finding the

low-dimensional representation, the hashing functions are finallyobtained through multivariable logistic regression. The proposedmethod is systematically evaluated on three datasets: Caltech-and the results show that ourmethod significantly outperforms the state-of-the-art multiviewhashing techniques.