Research And Application Of Image Retrieval Algorithm Based On Content And Its System

In recent years,with the rapid development of computer software and hardware technology and Internet communication technology,as well as digital cameras,smart phones,tablet PCs and other large-capacity storage devices and mobile terminal equipment increasingly cheap and popular,people can be very easy to store or browse A large number of digital image resources.And in people's work,life and entertainment and other fields are widely used in the image data,as an important carrier for the acquisition and exchange of information,showing an unprecedented massive explosive rapid growth trend.The essence of image retrieval is the extraction of image features and the matching of image features.Therefore,the image feature needs to reflect the semantic information of the image,which can be divided into shallow image feature,middle image feature and deep image feature.In the practical application,we can choose the appropriate image feature according to different needs.In order to satisfy the accuracy and real-time performance of the system response of the user in the image retrieval when the amount of image data is large or the dimension is high,the content-based image retrieval system generally uses a method based on the image hash method.Therefore,this thesis focuses on the research and application of image retrieval based on hash coding.The main work and contribution of the thesis are summarized as follows: Firstly,an improved iterative quantization hash algorithm is proposed to realize the image retrieval,the derivation process of the algorithm is described,and the algorithm model is established.Based on the existing iterative quantization hash algorithm,this algorithm uses the NMF algorithm instead of the PCA algorithm to reduce the image feature and effectively express the local features of the image.At the same time,we introduce the sparse reconstruction of the feature matrix of the graph,so that the improved algorithm can capture the structural information of the sample.Then,the simulation experiments are carried out on two kinds of open data sets,and the improved algorithm is compared with the popular unsupervised hash algorithm.The experimental results show that the recall rate,accuracy and average retrieval accuracy of the improved algorithm are better than the contrasted image hash algorithm.Parameter sensitivity experiments show that the algorithm is not sensitive to the parameters.This thesis also compares the running time of different algorithms,and finally proposes a simple way to select the hash code bits.Finally,this thesis applies the algorithm to the image management subsystem in the enterprise management system,and analyzes the result and performance parameters of the algorithm.