Research On The Method Of Region-Based Image Retrieval

With the rapid develooment of modern information technology,it will need the better image retrieval to search the massive image data.So that the content-based image information retrieval is proposed and developed rapidly.It contains the image retrieval based on global feature and image retrieval based on provincial characteristics.The former method uses the global visual features to match the image directly.The latter method is that it divided the image into a number of target area at first,and then used the matching algorithm based on provincial characteristics to search image.Im my subject,it mainly researched the critical technology for region division and region matching according to the image retrieval base on provincial characteristics.The subject put forward the method,multiscale segmentation of color image features which match the ISODATA clustering with the region growth algorithm.The method combines each characteristics of the clustering algorithm and the region growing algorithm to select equably the cpoint seed for ISODATA clustering by using 8*8 and 16*16 blocks of two sub-blocks of various scales of color,texture and shape characteristics and constructs growth criteria for segmentation of two scales based on region-growing,then merge the region segmentation for the two sub-blocks by using the location of the region and the characteristics of the region.So that it will become the large significant target area.In the process of regional matching,it used the characteristics of the mean of three features in the same type of regional block to carry out the calculation of the regional similarity.In order to better describe the semantic relationship between the image area.The subject improved the IRM algorithm of the importance of factors based on the regional distribution,the regional size and the color contrast between adjacent regions of factors and protruded the importance of image in the target area.The article calculated the similarity matrix after optimization of the regional characteristics by the improved DPF feature selection function and then combined improved IRM region matching algorithm to complete the regional match.According to three different researching method,we can find the method my used in the subject is better than the other methods which improves the precision of image retrieval.It also can achieve a better match result for the main areas of image matching.Especially,it will have the more accurate search results for the main large area of the image which have the color contrast with large.