Research On Deep-hashing Based Image Instance Retrieval Algorithm

With the development of big data and artificial intelligence technology,people need to search for interesting information from massive images.Therefore,hashing-based methods have been widely applied in large-scale image retrieval.Hashing methods aim to process images by projecting the feature space to the Hamming space with a similarity-preserving manner.Then retrieving images can be solved by calculating the Hamming distances of their corresponding hash codes,which can improve the storage efficiency and search performance.In recent years,as the convolutional neural networks has gained striking achievement.Due to the ability of feature extraction and end-to-end optimization,the deep-hashing methods have shown the superior performance to the traditional hashing methods.In this paper,we study deep-hashing based instance location and retrieval.First,the features of images are extracted by using the deep convolution networks aggregation method,and the gradient of feature density method is employed to locate the interested regions of images.On this basis,we proposed the deep-hashing based image instance retrieval method,where the 1?1 convolution kernel is incorporated into the optimization of the objective function.Then the hash codes of instances are generated to improve the performance of image retrieval by providing the fine-grained image searching.Finally,the experimental results on the CIFAR-10,VOC2012 and MSRCv2 datasets show that the proposed deep-hash based image instance retrieval method has a significant improvement in search accuracy compared to the traditional search algorithm.