Research Of Hashing-based Method For Image Retrieval

In recent years,with the rapid development of digital information related technologies,various types of image data have been increasing day by day,and a large number of images have been generated on the Internet.How to perform efficient image retrieval in a massive image database has become a huge challenge.Traditional image retrieval techniques,such as linear scanning and tree-based index structure,can achieve better results when the amount of data is small and the dimensions are low,but in the case of massive and high-dimensional data,their performance will become very poor and will not adapt to the search requirements.In this context,hash-based image retrieval technology emerges as the times require.The idea is to convert the original image into a compact binary code representation,which not only can greatly reduce the storage space,but also greatly speed up the retrieval speed.The hash-based image retrieval method can be better applied to retrieval in large-scale image databases.The process of hashing an image can be roughly divided into two stages of projection and quantization,which are crucial for the final image hash coding performance,in order to obtain highly robust hash coding to improve the efficiency of image retrieval,this paper has carried out in-depth research on the above two points.In order to ensure the performance of image hash coding,in the projection stage,for the redundancy between some features in the original feature space,the data is analyzed by the principal component analysis algorithm to project into a new feature subspace;in the quantization stage,taking into account the shortcomings of traditional quantization methods,in order to minimize the loss caused by quantization,we proposed a new quantization strategy which combine single and double bit quantization method.Based on this,combined with K-means clustering algorithm,the adaptive bit allocation quantization method is further proposed.The experimental results on the current popular image datasets show that the proposed method is superior to the existing mainstream image hashing algorithms under the multiple evaluation indexes such as accuracy and recall rate.