The Research On Content-based Image Retrieval

With the development of multimedia technology,an increasing number of people are keening to use images to record information in life,which also has led to an increase in the number of image data worldwide.Therefore,it is still a hot academic issue that how to search these massive image resources effectively for researchers.Image retrieval technology has developed rapidly in the past few decades,and more and more image retrieval methods have been proposed and improved.From text-based image retrieval technology to content-based image retrieval technology,people get rid of complicated manual image annotation,and they use effective image features to represent the content of images to achieve retrieval of specific images;from traditional manual feature design to learning-based image retrieval feature extraction methods,people's understanding of image features have changed from low-level colors,textures and shapes to high-level features which can represent image semantic content;from the continuous image feature vector to the discrete feature vector,people combine the hash technique with the deep learning to obtain the hash code that can fully represent the semantic content of the image,which greatly speeds up the image retrieval.This paper first introduces the content-based image retrieval technology,expounds some important achievements in the field of image retrieval in recent years,and deeply analyzes the key technologies.Next,the paper discusses the basic principle of image retrieval technology based on deep hash,and proposes its own image retrieval method based on multi-information flow and image retrieval method based on hybrid pooling model.Then,related experiments are carried out on MNIST,CIFRA-10 and NUS-WIDE,and experimental result demonstrates the effectiveness of the proposed method.Finally,this paper summarizes our contributions and looks forward to the future research focus.