The Research Of Image Retrieval Technology Based On Hashing Methods

With rapid increase of the image dataset scale,how to retrieve an image quickly and correctly has become an important research area in intelligent information processing.According to image features,image retrieval includes global feature based retrieval and local feature based retrieval.In local feature based retrieval,BOVW is the main method of image content.Constructing visual vocabulary and quantifying local features in the BOVW model is a critical step.The traditional visual vocabulary have the problem of low discriminative and long running time and without space information.In this paper,new methods of generating bag of visual word model are proposed:1)A method of generating bag of visual word model based on supervised hashing with kernel function is proposed.Our method does not need to take clustering,but only needs to learn the binary hashing functions,which reduces the complexity and the time of constructing visual vocabularies greatly.In the process of constructing visual vocabularies,we learn the data-relevant hashing functions,which generate the visual vocabularies with higher discriminability and does not need not generate multiple visual vocabularies to aggregate,but can achieve better retrieval performance than the existing methods.2)A method of generating bag of visual word model based on supervised hashing with kernel function and space pyramid is proposed.In order to consider the space information of the local features,we combine the space pyramid with the visual vocabularies and construct a visual vocabulary with stronger expression ability.Experimental results on different datasets show that the methods we proposed have higher classification accuracy which compared with other classifier methods.And the image retrieval performance of our methods are best which compared with other retrieval methods based on hashing.