Research On Nearest Neighbor Retrieval Of Large-scale Image Based On Hashing Method

The advent of Internet has resulted in massive information overloading in the recent decades.Due to the daramatic increase in the size of the data,modern information technology has to deal with such gigantic databases.In fact,compared to the cost of storage,searching for rich media data,such as audio,images and videos remains a major challenge since there exist major gaps between available solutions and pratical needs in both accuracy and computation cost.Hashing has become a popular and efficient technique foe efficient large-scale image and vedio retrieval recently.Hashing methods aim to map the original high-dimensional feature to compact binary code and preserve the semantic structure of the original feature in the Hamming space simultaneously.It turns out that using compact binary codes is much more efficient due to the employment of extremely fast Hamming distance computation.Our work focus on the challage of the hashing method,and propose novel hashing framework.Perserving the pairwise similarity of data points in the Hamming space is critical in state-of-the-art hashing techniques.However,most previous methods ignore to capture the local geometric structure residing on original data,which is essential for similarity search.In this work,we propose a novel hashing framework,which simultaneously optimizes similarity preserving hash codes and reconstructs the locally linear structures of data in the Hamming space.In specific,we learn two hash functions such that the resulting two sets of binary codes can well preserve the pairwise similarity and sparse neighborhood in the original feature space.We evaluate the proposed method on several large-scale image datasets,and the results demonstrate it significantly outperforms recent state-of-the-art hashing methods on large-scale image retrieval tasks.Deep neural network based hashing methods have greatly improved the multimedia retrieval performance by simutultaneously learning feature representations and binary hash functions.Inspired by the latest advance in the asymmetric hashing schme,in this work,we propose a novel deep hashing framework for supervised hashing.The core code is that two deep convolutional model are jointly trained such that their output codes for a pair of images can well reveal the similarity indicated by their semantic labels.By taking advantage of the exibility of asymmetric hash functions,we devise an efficient alternating algorithm to optimize the asymmetric deep hashfunctions and high-quality binary code jointly.Experiments on three image benchmarks show that DAPH achieves the state-of-the-art performance on large-scale image retrieval.