Research On Image Retrieval Method Of Combining SIFT Feature With PGH

With the development of network and multimedia technology, digital image hasbecome an important way for people to access information because of its rich content. Ithas been widely used in various fields for the intuitive and effective information formsof expression. How to retrieval the images you want quickly and accurately in a giganticimage database have become an urgent problem.The traditional text-based image retrieval method has been unable to meet people’sneeds. Content-based image retrieval has been developing over the past decade.Recently, the image color, texture, shape and other characteristics are commonly used incontent-based image retrieval method. However, the retrieval speed and accuracy cannot meet the user requirements. Image point features which contain image structuralinformation have been widely used in image matching, object detection, etc. In thispaper, the description method of the image point features based on the scale invariantfeature transform (SIFT) is researched in detail. An image retrieval method ofcombining SIFT features with pairwise geometric histogram (PGH) is proposed. Themain contents are as follows:(1) The image retrieval key technologies are researched and reviewed. The SIFTfeature extraction and PGH methods are studied in detail.(2) An image retrieval method of combining SIFT features with PGH is proposed.Firstly, the image SIFT local features are extracted to form SIFT descriptors. Then, thefeature points are used to extract the image convex hull to remove image redundantinformation. and form the PGH containing spatial distribution information. Finally,image similarity is measure by combining the SIFT features with PGH. Theexperimental results show that the image retrieval method of combining SIFT featureswith PGH can improved the precision and recall because image point features andpairwise geometric attributes are used.(3) a small prototype image retrieval system is designed and implemented based onVisua1C++and OpenCV. The proposed methods are realized combined with relevancefeedback mechanism. The basic image retrieval function is achieved.