Research On Content-based Image Retrieval Algorithm

With the popularization and application of Internet information technology,more and more image resources are uploaded to the network platform.For such a huge image data,people urgently need a technology to implement image retrieval and inquiries.Traditional image retrieval using annotation,due to the complexity of the process and lack of objectivity,it cannot adapt to current development trends,so content-based image retrieval has emerged.The content-based image retrieval technology can automatically analyze the expression information of the image,thus avoiding the human interference factors when the image is text-labeled.In order to improve the performance of image retrieval system,this paper focuses on the image sub-block weighted retrieval algorithm and the retrieval algorithm for the region of interest of the image based on the underlying features of the image.The specific research contents are as follows:(1)This article first discusses the analysis of the series of related content in the field of image retrieval.At the same time,it makes a preliminary analysis of the underlying features of the image,and then introduces in detail the feature similarity measurement methods that are commonly used in the search and is used to optimize the search results.In this paper,based on the research of series of related algorithms,two improved retrieval algorithms are proposed,which are weighted retrieval algorithm based on image segmentation and retrieval algorithm based on image region of interest.(2)The global feature of the image extracted by the traditional algorithm neglects the spatial distribution of the image.In view of this deficiency,a new image block weighting algorithm is proposed in the paper.Firstly,Harris interest points are extracted from the image.Then the image is divided into 3×3 subblocks of the same size and the number of interest points located in different subblock areas is counted.According to the interest points in each subblock area,the proportion of numbers is given to the corresponding weights of the blocks.Finally,the weighted feature vectors of each block are extracted and the feature similarity is compared.Experimental results show that the image block weighting algorithm proposed in this paper can more rationally reflect the spatial characteristics between regions and obtain better retrieval performance.(3)Apart from the main part,the image usually contains a large amount of background information.These redundant contents will undoubtedly affect the search results.If only extracting the regional features containing the subject of the image,this interference factor can be weakened and the accuracy of the search can be improved.Therefore,an improved region-of-interest extraction algorithm is proposed in this paper.Harris-Laplace algorithm is used to extract the interest points of the image and eliminate some interest points that are free in the edge of the image.Thus,an circumscribed convex polygon region based on interest points is obtained.This region is used as the region of interest of the image and only the information features in the region are extracted to achieve the search.Experimental results show that the improved algorithm proposed in this paper can extract regions of interest that are more in line with the human eye's perception,and achieve better retrieval results through the fusion of multiple features.