Research And Application Of Image Retrieval Based On Fast Search And Find Density Peak Clustering

With the advancement of human society and the development of hi-tech such as computer,Internet and storage technology,hundreds of millions of images are produced every day and propagated through various channels,so the number of digital images is increasing at a breathtaking pace.At present,what really matters is how to effectively manage and retrieve information from such a large image data to find the hidden values.Thus,we need an image retrieval technique which can query images and related information more quickly and accurate.Image clustering provides a new technical support for image retrieval.Based on clustering,images retrieval can explore the information which users are interested in from many image data quickly and accurately.However,the normal feature extraction algorithms applied to image clustering often neglect the spatial distribution of image color,and the adaptability is poor.This paper put forward an improved algorithm based on the DP clustering(Fast Search Density Peak),and apply it in the image retrieval system.The new algorithm divide the image through some equal area rectangular rings,and calculate the correlation and importance of each spatial region to merge the spatial information with color information.The new algorithm can improve the accuracy of cluster and ensure the convergence speed at same time.The main contents and work of this paper are as follows:(1)Study the color feature extraction and quantification methods.Most color spaces are proposed from the point of hardware,which cannot match with human eyes very well.This paper selects HSV color space as the color space model.Meanwhile,we quantitate visual perception of color to improve the speed of operation and make it easier for statistics and calculation.(2)Study the regional correlation calculation method.Traditional color feature extraction methods only input the color value to statistical analysis,and cannot take spatial distributions into account.To make the color feature more representative,this paper put forward a correlation calculation method based on image content,which can combine the spatial information with the color feature,and adjust the importance weights of each region automatically.The experiment results show that our algorithm can improve the robustness and universality of feature extraction.(3)Study the optimization scheme of DP clustering algorithm.In the original DP clustering algorithm,the truncation distance is fixed.However,this parameter determines the effect of clustering algorithm in a certain sense.Therefore,a suitable truncation distance has a significant effect on the clustering effect of DP algorithm.In this paper,we propose a dynamic adjustment scheme of truncation distance to ensure the algorithm has a faster convergence rate and higher clustering accuracy.The experiment results show that our method perform feasibly and effectively.